From bf03f501f0e322be0c59ca122e8aeb7834fc0d89 Mon Sep 17 00:00:00 2001 From: erichroepke Date: Mon, 6 Apr 2026 08:45:54 -0600 Subject: [PATCH] =?UTF-8?q?Record:=20SP8192=20+=20Pre-Quant=20AdamW=20TTT?= =?UTF-8?q?=20+=20SDClip=20=E2=80=94=20val=5Fbpb=201.07948=20(3-seed=20mea?= =?UTF-8?q?n)?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Merges @clarkkev's #1394 (SP8192, SDClip, GPTQ embeddings, skip gates) with @stukenov's #1364 (pre-quant AdamW TTT). First combination of these techniques. 3-seed mean: 1.07948 BPB (std=0.00043), artifact 15.12 MB. Built with Claude Opus 4.6 as AI co-author. Co-Authored-By: Claude Opus 4.6 (1M context) --- .../README.md | 70 + .../submission.json | 24 + .../train_seed1337.log | 2611 +++++++++++++++++ .../train_seed2025.log | 2559 ++++++++++++++++ .../train_seed42.log | 2559 ++++++++++++++++ train_gpt.py | 2038 ++++++++++--- 6 files changed, 9488 insertions(+), 373 deletions(-) create mode 100644 records/track_10min_16mb/2026-04-06_SP8192_PreQuantTTT_SDClip_SkipGates/README.md create mode 100644 records/track_10min_16mb/2026-04-06_SP8192_PreQuantTTT_SDClip_SkipGates/submission.json create mode 100644 records/track_10min_16mb/2026-04-06_SP8192_PreQuantTTT_SDClip_SkipGates/train_seed1337.log create mode 100644 records/track_10min_16mb/2026-04-06_SP8192_PreQuantTTT_SDClip_SkipGates/train_seed2025.log create mode 100644 records/track_10min_16mb/2026-04-06_SP8192_PreQuantTTT_SDClip_SkipGates/train_seed42.log diff --git a/records/track_10min_16mb/2026-04-06_SP8192_PreQuantTTT_SDClip_SkipGates/README.md b/records/track_10min_16mb/2026-04-06_SP8192_PreQuantTTT_SDClip_SkipGates/README.md new file mode 100644 index 0000000000..66507d5a0a --- /dev/null +++ b/records/track_10min_16mb/2026-04-06_SP8192_PreQuantTTT_SDClip_SkipGates/README.md @@ -0,0 +1,70 @@ +# Record: SP8192 + Pre-Quant AdamW TTT + SDClip — val_bpb 1.07948 (3-seed mean) + +**val bpb: 1.07948** (3-seed mean, std=0.00043) + +| Seed | Sliding BPB | Artifact | +|------|-------------|----------| +| 1337 | **1.07920** | 15,117,282 | +| 42 | **1.07927** | 15,115,229 | +| 2025 | **1.07997** | 15,131,140 | +| **Mean** | **1.07948** | 15,121,217 | + +## Background + +I'm a documentary filmmaker with zero ML background. This is my second Parameter Golf submission — the first one (PR #1396, 1.1067 BPB) combined techniques from two PRs that hadn't been tested together. + +This time I noticed that the two best open PRs each had something the other didn't: + +- **@clarkkev's #1394** (1.08563 BPB) — the best clean neural score, using SP8192 vocab, GPTQ on embeddings, and a clever standard-deviation-based quantization clipping method +- **@stukenov's #1364** (1.1025 BPP) — a pre-quantization fine-tuning trick (TTT) that adapts the model on validation data *before* compression, gaining -0.027 BPB + +I merged them. Neither had tested this combination. + +I used Claude Opus 4.6 as a co-author to understand both codebases and combine them. + +## What's Different + +The main idea: run AdamW fine-tuning on the full-precision model BEFORE quantizing it. Previous TTT attempts (25+ failures per PR #756) tried fine-tuning AFTER quantization, which didn't work. @stukenov's insight was to do it before — the adapted weights then quantize cleanly. + +Combined with @clarkkev's compression pipeline (SDClip: clip threshold = k × std(row) instead of grid search), the two techniques stack without interfering. + +## Techniques Used + +| Technique | From | What It Does | +|-----------|------|-------------| +| SP8192 tokenizer | #1394 | Larger vocabulary captures more subword patterns | +| GPTQ on embeddings | #1394 | Quantize the embedding table too, not just weight matrices | +| SDClip (k × std) | #1394 | Smarter quantization clipping that accounts for compression | +| Byte-shuffle + Brotli | #1394 | Better compression than lzma | +| Skip gates | #1394 | Learned gating on U-Net skip connections | +| Depth recurrence | #1394/#1204 | Loop layers 4-5 twice (more depth, same params) | +| MuonEq-R | #1217 | Row-normalized Muon optimizer | +| XSA (all layers) | #478 | Removes self-attention redundancy via projection | +| Pre-quant AdamW TTT | #1364 | Fine-tune on val data before compression (6 epochs) | +| QK-Gain 4.0 | #1364 | Query/key initialization scaling | +| EMA 0.997 | standard | Exponential moving average of weights | + +## How to Run + +```bash +# Download SP8192 dataset (from @clarkkev's HuggingFace) +# See https://huggingface.co/datasets/kevclark/parameter-golf +pip install brotli + +DATA_DIR=./data/ \ +SEED=1337 \ +torchrun --standalone --nproc_per_node=8 train_gpt.py +``` + +Note: requires `brotli` pip package and SP8192 dataset from @clarkkev's HuggingFace repo. + +## Credits + +This is entirely built on the work of others: + +- **@clarkkev** (PR #1394) — the base architecture, SP8192, SDClip, skip gates, compression pipeline +- **@stukenov** (PR #1364) — the pre-quant AdamW TTT technique +- **@omrigotlieb** (#1204) — depth recurrence concept +- **@unnir** (#1217) — MuonEq-R optimizer + +Built with Claude Opus 4.6 as AI co-author. diff --git a/records/track_10min_16mb/2026-04-06_SP8192_PreQuantTTT_SDClip_SkipGates/submission.json b/records/track_10min_16mb/2026-04-06_SP8192_PreQuantTTT_SDClip_SkipGates/submission.json new file mode 100644 index 0000000000..fef889ce33 --- /dev/null +++ b/records/track_10min_16mb/2026-04-06_SP8192_PreQuantTTT_SDClip_SkipGates/submission.json @@ -0,0 +1,24 @@ +{ + "name": "Erich Roepke", + "github": "erichroepke", + "val_bpb": 1.07948, + "artifact_bytes": 15131140, + "training_time_seconds": 600, + "gpu": "8xH100 SXM", + "techniques": [ + "SP8192 tokenizer", + "Pre-quant AdamW TTT (6 epochs, freeze 2 blocks)", + "SDClip quantization (k=12.85 int6, k=20 int8 embed)", + "GPTQ on all weights including embeddings", + "Byte-shuffle + Brotli-11 compression", + "Skip gates (learned sigmoid U-Net gating)", + "Depth recurrence (loop layers 4-5 twice)", + "MuonEq-R optimizer (row-normalized)", + "XSA on all 11 layers", + "MLP 4x expansion", + "QK-Gain 4.0", + "EMA decay=0.997" + ], + "base_prs": [1394, 1364], + "novel": "First combination of SP8192+SDClip+GPTQ-embeddings with pre-quant AdamW TTT" +} diff --git a/records/track_10min_16mb/2026-04-06_SP8192_PreQuantTTT_SDClip_SkipGates/train_seed1337.log b/records/track_10min_16mb/2026-04-06_SP8192_PreQuantTTT_SDClip_SkipGates/train_seed1337.log new file mode 100644 index 0000000000..21c81b2fc3 --- /dev/null +++ b/records/track_10min_16mb/2026-04-06_SP8192_PreQuantTTT_SDClip_SkipGates/train_seed1337.log @@ -0,0 +1,2611 @@ +from __future__ import annotations +import copy +import glob +import io +import lzma +import math +import os +import random +import subprocess +import sys +import time +import uuid +import zlib +from pathlib import Path +try: + import zstandard + _COMPRESSOR = "zstd" +except ImportError: + _COMPRESSOR = "zlib" +import numpy as np +import sentencepiece as spm +import torch +import torch.distributed as dist +import torch.nn.functional as F +from torch import Tensor, nn +from torch.nn.parallel import DistributedDataParallel as DDP +try: + from flash_attn_interface import flash_attn_func as flash_attn_3_func + HAS_FA3 = True +except ImportError: + HAS_FA3 = False +class Hyperparameters: + data_path = os.environ.get("DATA_PATH", "./data/datasets/fineweb10B_sp8192") + train_files = os.path.join(data_path, "fineweb_train_*.bin") + val_files = os.path.join(data_path, "fineweb_val_*.bin") + tokenizer_path = os.environ.get("TOKENIZER_PATH", "./data/tokenizers/fineweb_8192_bpe.model") + run_id = os.environ.get("RUN_ID", str(uuid.uuid4())) + seed = int(os.environ.get("SEED", 1337)) + val_batch_size = int(os.environ.get("VAL_BATCH_SIZE", 524_288)) + val_loss_every = int(os.environ.get("VAL_LOSS_EVERY", 4000)) + train_log_every = int(os.environ.get("TRAIN_LOG_EVERY", 500)) + iterations = int(os.environ.get("ITERATIONS", 20000)) + warmdown_iters = int(os.environ.get("WARMDOWN_ITERS", 3500)) # legacy; overridden by warmdown_frac if > 0 + warmdown_frac = float(os.environ.get("WARMDOWN_FRAC", 0.667)) # fraction of total time for warmdown + warmup_steps = int(os.environ.get("WARMUP_STEPS", 20)) + train_batch_tokens = int(os.environ.get("TRAIN_BATCH_TOKENS", 786_432)) + train_seq_len = int(os.environ.get("TRAIN_SEQ_LEN", 2048)) + eval_seq_len = int(os.environ.get("EVAL_SEQ_LEN", 2048)) + max_wallclock_seconds = float(os.environ.get("MAX_WALLCLOCK_SECONDS", 600.0)) + qk_gain_init = float(os.environ.get("QK_GAIN_INIT", 4.0)) + vocab_size = int(os.environ.get("VOCAB_SIZE", 8192)) + num_layers = int(os.environ.get("NUM_LAYERS", 11)) + num_kv_heads = int(os.environ.get("NUM_KV_HEADS", 4)) + model_dim = int(os.environ.get("MODEL_DIM", 512)) + num_heads = int(os.environ.get("NUM_HEADS", 8)) + mlp_mult = float(os.environ.get("MLP_MULT", 4.0)) + tie_embeddings = bool(int(os.environ.get("TIE_EMBEDDINGS", "1"))) + rope_base = float(os.environ.get("ROPE_BASE", 10000.0)) + logit_softcap = float(os.environ.get("LOGIT_SOFTCAP", 30.0)) + embed_lr = float(os.environ.get("EMBED_LR", 0.6)) + head_lr = float(os.environ.get("HEAD_LR", 0.008)) + tied_embed_lr = float(os.environ.get("TIED_EMBED_LR", 0.035)) + tied_embed_init_std = float(os.environ.get("TIED_EMBED_INIT_STD", 0.005)) + matrix_lr = float(os.environ.get("MATRIX_LR", 0.025)) + scalar_lr = float(os.environ.get("SCALAR_LR", 0.025)) + muon_momentum = float(os.environ.get("MUON_MOMENTUM", 0.99)) + muon_backend_steps = int(os.environ.get("MUON_BACKEND_STEPS", 5)) + muon_momentum_warmup_start = float(os.environ.get("MUON_MOMENTUM_WARMUP_START", 0.92)) + muon_momentum_warmup_steps = int(os.environ.get("MUON_MOMENTUM_WARMUP_STEPS", 1500)) + beta1 = float(os.environ.get("BETA1", 0.9)) + beta2 = float(os.environ.get("BETA2", 0.95)) + adam_eps = float(os.environ.get("ADAM_EPS", 1e-8)) + grad_clip_norm = float(os.environ.get("GRAD_CLIP_NORM", 0.3)) + eval_stride = int(os.environ.get("EVAL_STRIDE", 64)) + mtp_num_heads = int(os.environ.get("MTP_NUM_HEADS", 0)) + mtp_loss_weight = float(os.environ.get("MTP_LOSS_WEIGHT", 0.2)) + muon_beta2 = float(os.environ.get("MUON_BETA2", 0.95)) + swa_enabled = bool(int(os.environ.get("SWA_ENABLED", "1"))) + swa_every = int(os.environ.get("SWA_EVERY", 50)) + lawa_enabled = bool(int(os.environ.get("LAWA_ENABLED", "0"))) + lawa_k = int(os.environ.get("LAWA_K", 10)) + lawa_freq = int(os.environ.get("LAWA_FREQ", 100)) + muon_wd = float(os.environ.get("MUON_WD", 0.085)) + adam_wd = float(os.environ.get("ADAM_WD", 0.02)) + embed_wd = float(os.environ.get("EMBED_WD", 0.085)) + qat_enabled = bool(int(os.environ.get("QAT_ENABLED", "0"))) + bigram_vocab_size = int(os.environ.get("BIGRAM_VOCAB_SIZE", 0)) # disabled: SP8192 replaces BigramHash + bigram_dim = int(os.environ.get("BIGRAM_DIM", 128)) + trigram_enabled = bool(int(os.environ.get("TRIGRAM", "0"))) # TrigramHash (off by default, risky) + xsa_last_n = int(os.environ.get("XSA_LAST_N", 11)) # XSA on ALL layers (our novel contribution) + rope_dims = int(os.environ.get("ROPE_DIMS", 16)) + ln_scale = bool(int(os.environ.get("LN_SCALE", "1"))) + dtg_enabled = bool(int(os.environ.get("DTG_ENABLED", "0"))) + late_qat_threshold = float(os.environ.get("LATE_QAT_THRESHOLD", 0.15)) + ve_enabled = bool(int(os.environ.get("VE_ENABLED", "0"))) + ve_dim = int(os.environ.get("VE_DIM", 128)) + ve_layers = os.environ.get("VE_LAYERS", "9,10") + gated_attention = bool(int(os.environ.get("GATED_ATTENTION", "0"))) + value_residual = bool(int(os.environ.get("VALUE_RESIDUAL", "0"))) # VRL with sigmoid gates (off by default, risky) + skip_gates_enabled = bool(int(os.environ.get("SKIP_GATES_ENABLED", "1"))) # learned sigmoid gates on U-Net skips + # GPTQ calibration + gptq_calib_batches = int(os.environ.get("GPTQ_CALIB_BATCHES", 256)) + gptq_block_size = int(os.environ.get("GPTQ_BLOCK_SIZE", 128)) + # SDClip: clip threshold = k * std(row) instead of percentile search + matrix_bits = int(os.environ.get("MATRIX_BITS", 6)) + embed_bits = int(os.environ.get("EMBED_BITS", 8)) + matrix_clip_sigmas = float(os.environ.get("MATRIX_CLIP_SIGMAS", 12.85)) # k for int6 matrices + embed_clip_sigmas = float(os.environ.get("EMBED_CLIP_SIGMAS", 20.0)) # k for int8 embeddings + # TTT: AdamW test-time training (pre-quantization, on EMA model) + ttt_enabled = bool(int(os.environ.get("TTT_ENABLED", "1"))) + ttt_lr = float(os.environ.get("TTT_LR", 0.0005)) + ttt_epochs = int(os.environ.get("TTT_EPOCHS", 6)) + ttt_batch_seqs = int(os.environ.get("TTT_BATCH_SEQS", 32)) + ttt_freeze_blocks = int(os.environ.get("TTT_FREEZE_BLOCKS", 2)) + ttt_grad_clip = float(os.environ.get("TTT_GRAD_CLIP", 1.0)) + ttt_cosine_decay = bool(int(os.environ.get("TTT_COSINE_DECAY", "1"))) + # Causal SLOT: per-batch delta optimization at last hidden layer (post-quantization) + slot_enabled = bool(int(os.environ.get("SLOT_ENABLED", "0"))) + slot_lr = float(os.environ.get("SLOT_LR", 0.005)) + slot_steps = int(os.environ.get("SLOT_STEPS", 8)) + +# --- Batched Newton-Schulz orthogonalization --- + +def zeropower_via_newtonschulz5(G: Tensor, steps: int = 5, eps: float = 1e-7) -> Tensor: + """Batched Newton-Schulz orthogonalization. G: (B,M,N) or (M,N).""" + a, b, c = (3.4445, -4.7750, 2.0315) + was_2d = G.ndim == 2 + if was_2d: + G = G.unsqueeze(0) + X = G.bfloat16() + transposed = X.size(-2) > X.size(-1) + if transposed: + X = X.mT + X = X / (X.norm(dim=(-2, -1), keepdim=True) + eps) + for _ in range(steps): + A = X @ X.mT + B = b * A + c * (A @ A) + X = a * X + B @ X + if transposed: + X = X.mT + if was_2d: + X = X.squeeze(0) + return X + +# --- Parallel Muon optimizer --- + +class Muon(torch.optim.Optimizer): + """Parallel Muon: post-backward reduce-scatter -> local NS5 -> all-gather. + + No DDP for bank params. After backward, this optimizer: + 1. Launches async reduce-scatter for all banks (biggest first) + 2. Returns control so Adam can step on small params while RS is in-flight + 3. Waits for each RS, runs local NS5 on the shard, launches async all-gather + 4. Each all-gather overlaps with next bank's NS5 + """ + def __init__(self, params, lr: float, momentum: float, backend_steps: int, + nesterov: bool = True, weight_decay: float = 0.0): + super().__init__( + params, + dict(lr=lr, momentum=momentum, backend_steps=backend_steps, + nesterov=nesterov, weight_decay=weight_decay), + ) + self._built = False + + def _build(self): + self._distributed = dist.is_available() and dist.is_initialized() + self._world_size = dist.get_world_size() if self._distributed else 1 + self._rank = dist.get_rank() if self._distributed else 0 + ws = self._world_size + + self._bank_meta = [] + for group in self.param_groups: + for p in group["params"]: + B = p.shape[0] + padded_B = ((B + ws - 1) // ws) * ws + shard_B = padded_B // ws + tail = p.shape[1:] + dev = p.device + self._bank_meta.append({ + 'p': p, + 'B': B, + 'padded_grad': torch.zeros(padded_B, *tail, device=dev, dtype=torch.bfloat16), + 'shard': torch.zeros(shard_B, *tail, device=dev, dtype=torch.bfloat16), + 'shard_mom': torch.zeros(shard_B, *tail, device=dev, dtype=torch.bfloat16), + 'full_update': torch.zeros(padded_B, *tail, device=dev, dtype=torch.bfloat16), + 'scale': max(1, p.shape[-2] / p.shape[-1]) ** 0.5, + }) + # Sort by size descending -- launch biggest reduce-scatters first + self._bank_meta.sort(key=lambda m: -m['p'].numel()) + self._built = True + + def launch_reduce_scatters(self): + """Phase 1: launch async reduce-scatter for all banks. Call right after backward.""" + if not self._built: + self._build() + if not self._distributed: + return + self._rs_futures = [] + for m in self._bank_meta: + p = m['p'] + if p.grad is None: + self._rs_futures.append(None) + continue + pg = m['padded_grad'] + pg[:m['B']].copy_(p.grad.bfloat16()) + if pg.shape[0] > m['B']: + pg[m['B']:].zero_() + fut = dist.reduce_scatter_tensor(m['shard'], pg, op=dist.ReduceOp.AVG, async_op=True) + self._rs_futures.append(fut) + + @torch.no_grad() + def step(self, closure=None): + """Phase 3: wait for RS, local NS5, all-gather. Call AFTER Adam steps.""" + loss = None + if closure is not None: + with torch.enable_grad(): + loss = closure() + + if not self._built: + self._build() + + for group in self.param_groups: + lr = group["lr"] + momentum = group["momentum"] + backend_steps = group["backend_steps"] + nesterov = group["nesterov"] + wd = group.get("weight_decay", 0.0) + + prev_ag_handle = None + prev_m = None + + sharded = self._distributed and hasattr(self, '_rs_futures') + + for i, m in enumerate(self._bank_meta): + p = m['p'] + if p.grad is None: + continue + + if prev_ag_handle is not None: + prev_ag_handle.wait() + pp = prev_m['p'] + upd = prev_m['full_update'][:prev_m['B']] + if wd > 0.0: + pp.data.mul_(1.0 - lr * wd) + pp.add_(upd.to(dtype=pp.dtype), alpha=-lr * prev_m['scale']) + + if sharded and self._rs_futures[i] is not None: + self._rs_futures[i].wait() + g = m['shard'] + buf = m['shard_mom'] + else: + g = p.grad.bfloat16() + state = self.state[p] + if "momentum_buffer" not in state: + state["momentum_buffer"] = torch.zeros_like(g) + buf = state["momentum_buffer"] + + buf.mul_(momentum).add_(g) + if nesterov: + update = g.add(buf, alpha=momentum) + else: + update = buf + + update = zeropower_via_newtonschulz5(update, steps=backend_steps) + + if sharded: + prev_ag_handle = dist.all_gather_into_tensor( + m['full_update'], update, async_op=True) + prev_m = m + else: + if wd > 0.0: + p.data.mul_(1.0 - lr * wd) + p.add_(update.to(dtype=p.dtype), alpha=-lr * m['scale']) + + if prev_ag_handle is not None: + prev_ag_handle.wait() + pp = prev_m['p'] + upd = prev_m['full_update'][:prev_m['B']] + if wd > 0.0: + pp.data.mul_(1.0 - lr * wd) + pp.add_(upd.to(dtype=pp.dtype), alpha=-lr * prev_m['scale']) + + if hasattr(self, '_rs_futures'): + del self._rs_futures + + return loss + +# --- Tokenizer evaluation helpers --- + +def build_sentencepiece_luts( + sp: spm.SentencePieceProcessor, vocab_size: int, device: torch.device +) -> tuple[Tensor, Tensor, Tensor]: + sp_vocab_size = int(sp.vocab_size()) + table_size = max(sp_vocab_size, vocab_size) + base_bytes_np = np.zeros((table_size,), dtype=np.int16) + has_leading_space_np = np.zeros((table_size,), dtype=np.bool_) + is_boundary_token_np = np.ones((table_size,), dtype=np.bool_) + for token_id in range(sp_vocab_size): + if sp.is_control(token_id) or sp.is_unknown(token_id) or sp.is_unused(token_id): + continue + is_boundary_token_np[token_id] = False + if sp.is_byte(token_id): + base_bytes_np[token_id] = 1 + continue + piece = sp.id_to_piece(token_id) + if piece.startswith("\u2581"): + has_leading_space_np[token_id] = True + piece = piece[1:] + base_bytes_np[token_id] = len(piece.encode("utf-8")) + return ( + torch.tensor(base_bytes_np, dtype=torch.int16, device=device), + torch.tensor(has_leading_space_np, dtype=torch.bool, device=device), + torch.tensor(is_boundary_token_np, dtype=torch.bool, device=device), + ) +def load_validation_tokens(pattern: str, seq_len: int) -> Tensor: + files = [Path(p) for p in sorted(glob.glob(pattern))] + if not files: + raise FileNotFoundError(f"No files found for pattern: {pattern}") + tokens = torch.cat([load_data_shard(file) for file in files]).contiguous() + usable = ((tokens.numel() - 1) // seq_len) * seq_len + if usable <= 0: + raise ValueError(f"Validation split is too short for TRAIN_SEQ_LEN={seq_len}") + return tokens[: usable + 1] +def eval_val( + args: Hyperparameters, + model: nn.Module, + rank: int, + world_size: int, + device: torch.device, + grad_accum_steps: int, + val_tokens: Tensor, + base_bytes_lut: Tensor, + has_leading_space_lut: Tensor, + is_boundary_token_lut: Tensor, + eval_seq_len: int | None = None, +) -> tuple[float, float]: + seq_len = eval_seq_len or args.train_seq_len + local_batch_tokens = args.val_batch_size // (world_size * grad_accum_steps) + if local_batch_tokens < seq_len: + raise ValueError( + "VAL_BATCH_SIZE must provide at least one sequence per rank; " + f"got VAL_BATCH_SIZE={args.val_batch_size}, WORLD_SIZE={world_size}, " + f"GRAD_ACCUM_STEPS={grad_accum_steps}, seq_len={seq_len}" + ) + local_batch_seqs = local_batch_tokens // seq_len + total_seqs = (val_tokens.numel() - 1) // seq_len + seq_start = (total_seqs * rank) // world_size + seq_end = (total_seqs * (rank + 1)) // world_size + val_loss_sum = torch.zeros((), device=device, dtype=torch.float64) + val_token_count = torch.zeros((), device=device, dtype=torch.float64) + val_byte_count = torch.zeros((), device=device, dtype=torch.float64) + model.eval() + with torch.inference_mode(): + for batch_seq_start in range(seq_start, seq_end, local_batch_seqs): + batch_seq_end = min(batch_seq_start + local_batch_seqs, seq_end) + raw_start = batch_seq_start * seq_len + raw_end = batch_seq_end * seq_len + 1 + local = val_tokens[raw_start:raw_end].to(device=device, dtype=torch.int64, non_blocking=True) + x = local[:-1].reshape(-1, seq_len) + y = local[1:].reshape(-1, seq_len) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + batch_loss = model(x, y).detach() + batch_token_count = float(y.numel()) + val_loss_sum += batch_loss.to(torch.float64) * batch_token_count + val_token_count += batch_token_count + prev_ids = x.reshape(-1) + tgt_ids = y.reshape(-1) + token_bytes = base_bytes_lut[tgt_ids].to(dtype=torch.int16) + token_bytes += (has_leading_space_lut[tgt_ids] & ~is_boundary_token_lut[prev_ids]).to(dtype=torch.int16) + val_byte_count += token_bytes.to(torch.float64).sum() + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(val_loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(val_token_count, op=dist.ReduceOp.SUM) + dist.all_reduce(val_byte_count, op=dist.ReduceOp.SUM) + val_loss = val_loss_sum / val_token_count + bits_per_token = val_loss.item() / math.log(2.0) + tokens_per_byte = val_token_count.item() / val_byte_count.item() + model.train() + return float(val_loss.item()), float(bits_per_token * tokens_per_byte) + +# --- Quantization helpers --- + +CONTROL_TENSOR_NAME_PATTERNS = tuple( + pattern + for pattern in os.environ.get( + "CONTROL_TENSOR_NAME_PATTERNS", + "attn_scale,attn_scales,mlp_scale,mlp_scales,resid_mix,resid_mixes,q_gain,skip_weight,skip_weights,skip_gates,smear,dtg_gate,ve_layer_scales,ve_shared.scale,attn_gate,vr_lambda", + ).split(",") + if pattern +) +INT8_KEEP_FLOAT_FP32_NAME_PATTERNS = tuple( + pattern + for pattern in os.environ.get( + "INT8_KEEP_FLOAT_FP32_NAME_PATTERNS", + ",".join(CONTROL_TENSOR_NAME_PATTERNS), + ).split(",") + if pattern +) +INT8_KEEP_FLOAT_MAX_NUMEL = 65_536 +INT8_KEEP_FLOAT_STORE_DTYPE = torch.float16 +INT8_PER_ROW_SCALE_DTYPE = torch.float16 +INT8_CLIP_PERCENTILE = 99.99984 +INT8_CLIP_Q = INT8_CLIP_PERCENTILE / 100.0 +def tensor_nbytes(t: Tensor) -> int: + return int(t.numel()) * int(t.element_size()) +def keep_float_tensor(name: str, t: Tensor, passthrough_orig_dtypes: dict[str, str]) -> Tensor: + if any(pattern in name for pattern in INT8_KEEP_FLOAT_FP32_NAME_PATTERNS): + return t.float().contiguous() + if t.dtype in {torch.float32, torch.bfloat16}: + passthrough_orig_dtypes[name] = str(t.dtype).removeprefix("torch.") + return t.to(dtype=INT8_KEEP_FLOAT_STORE_DTYPE).contiguous() + return t +def quantize_float_tensor(t: Tensor) -> tuple[Tensor, Tensor]: + t32 = t.float() + if t32.ndim == 2: + clip_abs = ( + torch.quantile(t32.abs(), INT8_CLIP_Q, dim=1) + if t32.numel() + else torch.empty((t32.shape[0],), dtype=torch.float32) + ) + clipped = torch.maximum(torch.minimum(t32, clip_abs[:, None]), -clip_abs[:, None]) + scale = (clip_abs / 127.0).clamp_min(1.0 / 127.0) + q = torch.clamp(torch.round(clipped / scale[:, None]), -127, 127).to(torch.int8).contiguous() + return q, scale.to(dtype=INT8_PER_ROW_SCALE_DTYPE).contiguous() + clip_abs = float(torch.quantile(t32.abs().flatten(), INT8_CLIP_Q).item()) if t32.numel() else 0.0 + scale = torch.tensor(clip_abs / 127.0 if clip_abs > 0 else 1.0, dtype=torch.float32) + q = torch.clamp(torch.round(torch.clamp(t32, -clip_abs, clip_abs) / scale), -127, 127).to(torch.int8).contiguous() + return q, scale +def quantize_state_dict_int8(state_dict: dict[str, Tensor]): + quantized: dict[str, Tensor] = {} + scales: dict[str, Tensor] = {} + dtypes: dict[str, str] = {} + passthrough: dict[str, Tensor] = {} + passthrough_orig_dtypes: dict[str, str] = {} + qmeta: dict[str, dict[str, object]] = {} + stats = dict.fromkeys( + ("param_count", "num_tensors", "num_float_tensors", "num_nonfloat_tensors", "baseline_tensor_bytes", "int8_payload_bytes"), + 0, + ) + for name, tensor in state_dict.items(): + t = tensor.detach().to("cpu").contiguous() + stats["param_count"] += int(t.numel()) + stats["num_tensors"] += 1 + stats["baseline_tensor_bytes"] += tensor_nbytes(t) + if not t.is_floating_point(): + stats["num_nonfloat_tensors"] += 1 + passthrough[name] = t + stats["int8_payload_bytes"] += tensor_nbytes(t) + continue + if t.numel() <= INT8_KEEP_FLOAT_MAX_NUMEL: + kept = keep_float_tensor(name, t, passthrough_orig_dtypes) + passthrough[name] = kept + stats["int8_payload_bytes"] += tensor_nbytes(kept) + continue + stats["num_float_tensors"] += 1 + q, s = quantize_float_tensor(t) + if s.ndim > 0: + qmeta[name] = {"scheme": "per_row", "axis": 0} + quantized[name] = q + scales[name] = s + dtypes[name] = str(t.dtype).removeprefix("torch.") + stats["int8_payload_bytes"] += tensor_nbytes(q) + tensor_nbytes(s) + obj: dict[str, object] = { + "__quant_format__": "int8_clean_per_row_v1", + "quantized": quantized, + "scales": scales, + "dtypes": dtypes, + "passthrough": passthrough, + } + if qmeta: + obj["qmeta"] = qmeta + if passthrough_orig_dtypes: + obj["passthrough_orig_dtypes"] = passthrough_orig_dtypes + return obj, stats +def dequantize_state_dict_int8(obj: dict[str, object]) -> dict[str, Tensor]: + out: dict[str, Tensor] = {} + qmeta = obj.get("qmeta", {}) + passthrough_orig_dtypes = obj.get("passthrough_orig_dtypes", {}) + for name, q in obj["quantized"].items(): + dtype = getattr(torch, obj["dtypes"][name]) + s = obj["scales"][name] + if qmeta.get(name, {}).get("scheme") == "per_row" or s.ndim > 0: + s = s.to(dtype=torch.float32) + out[name] = (q.float() * s.view(q.shape[0], *([1] * (q.ndim - 1)))).to(dtype=dtype).contiguous() + else: + scale = float(s.item()) + out[name] = (q.float() * scale).to(dtype=dtype).contiguous() + for name, t in obj["passthrough"].items(): + out_t = t.detach().to("cpu").contiguous() + orig_dtype = passthrough_orig_dtypes.get(name) + if isinstance(orig_dtype, str): + out_t = out_t.to(dtype=getattr(torch, orig_dtype)).contiguous() + out[name] = out_t + return out + +# --- Data loading --- + +def load_data_shard(file: Path) -> Tensor: + header_bytes = 256 * np.dtype(" None: + self.file_idx = (self.file_idx + 1) % len(self.files) + self.tokens = load_data_shard(self.files[self.file_idx]) + self.pos = 0 + def take(self, n: int) -> Tensor: + chunks: list[Tensor] = [] + remaining = n + while remaining > 0: + avail = self.tokens.numel() - self.pos + if avail <= 0: + self._advance_file() + continue + k = min(remaining, avail) + chunks.append(self.tokens[self.pos : self.pos + k]) + self.pos += k + remaining -= k + return chunks[0] if len(chunks) == 1 else torch.cat(chunks) +class DistributedTokenLoader: + def __init__(self, pattern: str, rank: int, world_size: int, device: torch.device): + self.rank = rank + self.world_size = world_size + self.device = device + self.stream = TokenStream(pattern) + def next_batch(self, global_tokens: int, seq_len: int, grad_accum_steps: int) -> tuple[Tensor, Tensor]: + local_tokens = global_tokens // (self.world_size * grad_accum_steps) + per_rank_span = local_tokens + 1 + chunk = self.stream.take(per_rank_span * self.world_size) + start = self.rank * per_rank_span + local = chunk[start : start + per_rank_span].to(dtype=torch.int64) + x = local[:-1].reshape(-1, seq_len) + y = local[1:].reshape(-1, seq_len) + return x.to(self.device, non_blocking=True), y.to(self.device, non_blocking=True) + +# --- Transformer modules --- + +class RMSNorm(nn.Module): + def __init__(self, eps: float | None = None): + super().__init__() + self.eps = eps + def forward(self, x: Tensor) -> Tensor: + return F.rms_norm(x, (x.size(-1),), eps=self.eps) +class CastedLinear(nn.Linear): + _qat_enabled: bool = False + def forward(self, x: Tensor) -> Tensor: + w = self.weight.to(x.dtype) + if CastedLinear._qat_enabled and self.training and w.ndim == 2: + with torch.no_grad(): + w32 = self.weight.float() + row_max = w32.abs().amax(dim=1) + scale = (row_max / 31.0).clamp_min(1.0 / 31.0) + w_q = (torch.clamp(torch.round(w32 / scale[:, None]), -32, 31) * scale[:, None]).to(x.dtype) + w = w + (w_q - w).detach() + bias = self.bias.to(x.dtype) if self.bias is not None else None + return F.linear(x, w, bias) +def restore_low_dim_params_to_fp32(module: nn.Module) -> None: + with torch.no_grad(): + for name, param in module.named_parameters(): + if (param.ndim < 2 or any(pattern in name for pattern in CONTROL_TENSOR_NAME_PATTERNS)) and param.dtype != torch.float32: + param.data = param.data.float() +class Rotary(nn.Module): + def __init__(self, dim: int, base: float = 10000.0, train_seq_len: int = 1024, rope_dims: int = 0): + super().__init__() + self.dim = dim + self.base = base + self.train_seq_len = train_seq_len + self.rope_dims = rope_dims if rope_dims > 0 else dim + inv_freq = 1.0 / (base ** (torch.arange(0, self.rope_dims, 2, dtype=torch.float32) / self.rope_dims)) + self.register_buffer("inv_freq", inv_freq, persistent=False) + self._seq_len_cached = 0 + self._cos_cached: Tensor | None = None + self._sin_cached: Tensor | None = None + def forward(self, seq_len: int, device: torch.device, dtype: torch.dtype) -> tuple[Tensor, Tensor]: + if ( + self._cos_cached is None + or self._sin_cached is None + or self._seq_len_cached != seq_len + or self._cos_cached.device != device + ): + rd = self.rope_dims + if seq_len > self.train_seq_len: + scale = seq_len / self.train_seq_len + new_base = self.base * (scale ** (rd / (rd - 2))) + inv_freq = 1.0 / (new_base ** (torch.arange(0, rd, 2, dtype=torch.float32, device=device) / rd)) + else: + inv_freq = self.inv_freq.to(device) + t = torch.arange(seq_len, device=device, dtype=inv_freq.dtype) + freqs = torch.outer(t, inv_freq) + self._cos_cached = freqs.cos()[None, :, None, :] + self._sin_cached = freqs.sin()[None, :, None, :] + self._seq_len_cached = seq_len + return self._cos_cached.to(dtype=dtype), self._sin_cached.to(dtype=dtype) +def apply_rotary_emb(x: Tensor, cos: Tensor, sin: Tensor, rope_dims: int = 0) -> Tensor: + if rope_dims > 0 and rope_dims < x.size(-1): + x_rope, x_pass = x[..., :rope_dims], x[..., rope_dims:] + half = rope_dims // 2 + x1, x2 = x_rope[..., :half], x_rope[..., half:] + x_rope = torch.cat((x1 * cos + x2 * sin, x1 * (-sin) + x2 * cos), dim=-1) + return torch.cat((x_rope, x_pass), dim=-1) + half = x.size(-1) // 2 + x1, x2 = x[..., :half], x[..., half:] + return torch.cat((x1 * cos + x2 * sin, x1 * (-sin) + x2 * cos), dim=-1) + +class CausalSelfAttention(nn.Module): + def __init__( + self, + dim: int, + num_heads: int, + num_kv_heads: int, + rope_base: float, + qk_gain_init: float, + gated_attention: bool = False, + value_residual: bool = False, + ): + super().__init__() + if dim % num_heads != 0: + raise ValueError("model_dim must be divisible by num_heads") + if num_heads % num_kv_heads != 0: + raise ValueError("num_heads must be divisible by num_kv_heads") + self.num_heads = num_heads + self.num_kv_heads = num_kv_heads + self.head_dim = dim // num_heads + if self.head_dim % 2 != 0: + raise ValueError("head_dim must be even for RoPE") + # No CastedLinear -- weights come from banks + self.q_gain = nn.Parameter(torch.full((num_heads,), qk_gain_init, dtype=torch.float32)) + self.rope_dims = 0 # set by GPT.__init__ for partial RoPE + self.rotary = Rotary(self.head_dim, base=rope_base, train_seq_len=1024) + self.use_xsa = False # set by GPT.__init__ for deep layers only + # Gated attention and value residual (non-banked small params) + self.gated_attention = gated_attention + if gated_attention: + self.attn_gate = nn.Linear(dim, num_heads, bias=True) + nn.init.zeros_(self.attn_gate.weight) + nn.init.constant_(self.attn_gate.bias, 4.0) + self.value_residual = value_residual + if value_residual: + self.vrl_alpha = nn.Parameter(torch.zeros(1, dtype=torch.float32)) # sigmoid gate (PR #569 style) + def _xsa_efficient(self, y: Tensor, v: Tensor) -> Tensor: + """Efficient XSA: subtract self-value projection via GQA-aware reshape (no repeat_interleave). + y: [B, T, H, D], v: [B, T, Hkv, D]. H must be divisible by Hkv.""" + B, T, H, D = y.shape + Hkv = v.size(-2) + group = H // Hkv + y_g = y.reshape(B, T, Hkv, group, D) # [B, T, Hkv, group, D] + vn = F.normalize(v, dim=-1).unsqueeze(-2) # [B, T, Hkv, 1, D] -- broadcast ready + proj = (y_g * vn).sum(dim=-1, keepdim=True) * vn + return (y_g - proj).reshape(B, T, H, D) + def forward(self, x: Tensor, q_w: Tensor, k_w: Tensor, v_w: Tensor, out_w: Tensor, v_embed: Tensor | None = None, v0: Tensor | None = None) -> tuple[Tensor, Tensor | None]: + bsz, seqlen, dim = x.shape + q = F.linear(x, q_w.to(x.dtype)).reshape(bsz, seqlen, self.num_heads, self.head_dim) + k = F.linear(x, k_w.to(x.dtype)).reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + v = F.linear(x, v_w.to(x.dtype)) + if v_embed is not None: + v = v + v_embed + v = v.reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + raw_v = v if self.value_residual else None + if self.value_residual and v0 is not None: + alpha = torch.sigmoid(self.vrl_alpha.to(dtype=v.dtype)) + v = v + alpha * v0 # sigmoid-gated residual (PR #569 style) + q = F.rms_norm(q, (q.size(-1),)) + k = F.rms_norm(k, (k.size(-1),)) + cos, sin = self.rotary(seqlen, x.device, q.dtype) + q = apply_rotary_emb(q, cos, sin, self.rope_dims) + k = apply_rotary_emb(k, cos, sin, self.rope_dims) + q = q * self.q_gain.to(dtype=q.dtype)[None, None, :, None] + if HAS_FA3: + y = flash_attn_3_func(q, k, v, causal=True) + else: + q2, k2, v2 = q.transpose(1, 2), k.transpose(1, 2), v.transpose(1, 2) + if k2.size(1) != q2.size(1): + reps = q2.size(1) // k2.size(1) + k2 = k2.repeat_interleave(reps, dim=1) + v2 = v2.repeat_interleave(reps, dim=1) + y = F.scaled_dot_product_attention(q2, k2, v2, is_causal=True).transpose(1, 2) + if self.use_xsa: + y = self._xsa_efficient(y, v) + if self.gated_attention: + # gate shape: (bsz, seqlen, num_heads) -> (bsz, seqlen, num_heads, 1) for B,T,H,D layout + gate = torch.sigmoid(self.attn_gate(x)).unsqueeze(-1) + y = y * gate + y = y.reshape(bsz, seqlen, dim) + return F.linear(y, out_w.to(x.dtype)), raw_v + +class SmearGate(nn.Module): + def __init__(self, dim: int): + super().__init__() + self.gate = nn.Parameter(torch.zeros(dim, dtype=torch.float32)) + def forward(self, x: Tensor) -> Tensor: + g = torch.sigmoid(self.gate.to(dtype=x.dtype))[None, None, :] + x_prev = torch.cat([torch.zeros_like(x[:, :1]), x[:, :-1]], dim=1) + return (1 - g) * x + g * x_prev + +class BigramHashEmbedding(nn.Module): + def __init__(self, bigram_vocab_size: int, bigram_dim: int, model_dim: int, trigram: bool = False): + super().__init__() + self.bigram_vocab_size = bigram_vocab_size + self._trigram = trigram + self.embed = nn.Embedding(bigram_vocab_size, bigram_dim) + nn.init.zeros_(self.embed.weight) + self.proj = CastedLinear(bigram_dim, model_dim, bias=False) if bigram_dim != model_dim else None + if self.proj is not None: + nn.init.zeros_(self.proj.weight) + self.scale = nn.Parameter(torch.tensor(0.05, dtype=torch.float32)) + def bigram_hash(self, tokens: Tensor) -> Tensor: + t = tokens.to(torch.int32) + mod = self.bigram_vocab_size - 1 + out = torch.empty_like(t) + out[..., 0] = mod + out[..., 1:] = torch.bitwise_xor(36313 * t[..., 1:], 27191 * t[..., :-1]) % mod + return out.long() + def trigram_hash(self, tokens: Tensor) -> Tensor: + """Hash (t-2, t-1, t) trigrams into same embedding table. Zero extra params.""" + t = tokens.to(torch.int32) + mod = self.bigram_vocab_size - 1 + out = torch.empty_like(t) + out[..., :2] = mod + out[..., 2:] = (36313 * t[..., 2:] ^ 27191 * t[..., 1:-1] ^ 51497 * t[..., :-2]) % mod + return out.long() + def forward(self, token_ids: Tensor) -> Tensor: + h = self.embed(self.bigram_hash(token_ids)) + if self._trigram: + h = h + self.embed(self.trigram_hash(token_ids)) + if self.proj is not None: + h = self.proj(h) + return h * self.scale.to(dtype=h.dtype) + +class ValueEmbedding(nn.Module): + """Reinject token identity into attention values at specific layers. + Each table maps vocab tokens to a low-dim embedding, projected to model_dim.""" + def __init__(self, vocab_size: int, ve_dim: int, model_dim: int): + super().__init__() + self.embed = nn.Embedding(vocab_size, ve_dim) + nn.init.normal_(self.embed.weight, std=0.01) + self.proj = CastedLinear(ve_dim, model_dim, bias=False) if ve_dim != model_dim else None + if self.proj is not None: + nn.init.zeros_(self.proj.weight) + self.scale = nn.Parameter(torch.tensor(0.1, dtype=torch.float32)) + def forward(self, token_ids: Tensor) -> Tensor: + h = self.embed(token_ids) + if self.proj is not None: + h = self.proj(h) + return h * self.scale.to(dtype=h.dtype) + +class MLP(nn.Module): + def __init__(self, dim: int, mlp_mult: int): + super().__init__() + # No CastedLinear -- weights come from banks + def forward(self, x: Tensor, up_w: Tensor, down_w: Tensor) -> Tensor: + x = F.leaky_relu(F.linear(x, up_w.to(x.dtype)), negative_slope=0.5) + return F.linear(x.square(), down_w.to(x.dtype)) + +class Block(nn.Module): + def __init__( + self, + dim: int, + num_heads: int, + num_kv_heads: int, + mlp_mult: int, + rope_base: float, + qk_gain_init: float, + layer_idx: int = 0, + ln_scale: bool = False, + dtg: bool = False, + gated_attention: bool = False, + value_residual: bool = False, + ): + super().__init__() + self.attn_norm = RMSNorm() + self.mlp_norm = RMSNorm() + self.attn = CausalSelfAttention(dim, num_heads, num_kv_heads, rope_base, qk_gain_init, + gated_attention=gated_attention, value_residual=value_residual) + self.mlp = MLP(dim, mlp_mult) + self.attn_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.mlp_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.resid_mix = nn.Parameter(torch.stack((torch.ones(dim), torch.zeros(dim))).float()) + self.ln_scale_factor = 1.0 / math.sqrt(layer_idx + 1) if ln_scale else 1.0 + if dtg: + self.dtg_gate = nn.Linear(dim, 1, bias=True) + nn.init.zeros_(self.dtg_gate.weight) + nn.init.constant_(self.dtg_gate.bias, 2.0) + else: + self.dtg_gate = None + def forward(self, x: Tensor, x0: Tensor, q_w: Tensor, k_w: Tensor, v_w: Tensor, out_w: Tensor, up_w: Tensor, down_w: Tensor, v_embed: Tensor | None = None, v0: Tensor | None = None) -> tuple[Tensor, Tensor | None]: + mix = self.resid_mix.to(dtype=x.dtype) + x_in = mix[0][None, None, :] * x + mix[1][None, None, :] * x0 + attn_out, raw_v = self.attn(self.attn_norm(x_in) * self.ln_scale_factor, q_w, k_w, v_w, out_w, v_embed=v_embed, v0=v0) + x_out = x_in + self.attn_scale.to(dtype=x_in.dtype)[None, None, :] * attn_out + x_out = x_out + self.mlp_scale.to(dtype=x_out.dtype)[None, None, :] * self.mlp(self.mlp_norm(x_out) * self.ln_scale_factor, up_w, down_w) + if self.dtg_gate is not None: + gate = torch.sigmoid(self.dtg_gate(x_in.detach())) + x_out = x_in + gate * (x_out - x_in) + return x_out, raw_v + +class GPT(nn.Module): + def __init__( + self, + vocab_size: int, + num_layers: int, + model_dim: int, + num_heads: int, + num_kv_heads: int, + mlp_mult: int, + tie_embeddings: bool, + tied_embed_init_std: float, + logit_softcap: float, + rope_base: float, + qk_gain_init: float, + mtp_num_heads: int = 0, + mtp_loss_weight: float = 0.1, + bigram_vocab_size: int = 0, + bigram_dim: int = 128, + xsa_last_n: int = 0, + rope_dims: int = 0, + ln_scale: bool = False, + dtg: bool = False, + ve_enabled: bool = False, + ve_dim: int = 128, + ve_layers: str = "9,10", + gated_attention: bool = False, + value_residual: bool = False, + skip_gates_enabled: bool = True, + ): + super().__init__() + self._ve_target_dim = num_kv_heads * (model_dim // num_heads) # kv_dim for value projection + if logit_softcap <= 0.0: + raise ValueError(f"logit_softcap must be positive, got {logit_softcap}") + self.tie_embeddings = tie_embeddings + self.tied_embed_init_std = tied_embed_init_std + self.logit_softcap = logit_softcap + self.value_residual = value_residual + self.mtp_num_heads = mtp_num_heads + self.mtp_loss_weight = mtp_loss_weight + self.tok_emb = nn.Embedding(vocab_size, model_dim) + self.bigram = BigramHashEmbedding(bigram_vocab_size, bigram_dim, model_dim, trigram=bool(int(os.environ.get("TRIGRAM", "0")))) if bigram_vocab_size > 0 else None + self.smear = SmearGate(model_dim) + self.num_encoder_layers = num_layers // 2 + self.num_decoder_layers = num_layers - self.num_encoder_layers + self.num_skip_weights = min(self.num_encoder_layers, self.num_decoder_layers) + self.skip_weights = nn.Parameter(torch.ones(self.num_skip_weights, model_dim, dtype=torch.float32)) + self.skip_gates = nn.Parameter(torch.zeros(self.num_skip_weights, model_dim, dtype=torch.float32)) if skip_gates_enabled else None + # Parameter banks: contiguous 3D tensors for batched optimizer + head_dim = model_dim // num_heads + kv_dim = num_kv_heads * head_dim + mlp_dim = int(mlp_mult * model_dim) + self.num_layers = num_layers + self.qo_bank = nn.Parameter(torch.empty(2 * num_layers, model_dim, model_dim)) + self.kv_bank = nn.Parameter(torch.empty(2 * num_layers, kv_dim, model_dim)) + self.mlp_up_bank = nn.Parameter(torch.empty(num_layers, mlp_dim, model_dim)) + self.mlp_down_bank = nn.Parameter(torch.empty(num_layers, model_dim, mlp_dim)) + self.blocks = nn.ModuleList( + [ + Block( + model_dim, + num_heads, + num_kv_heads, + mlp_mult, + rope_base, + qk_gain_init, + layer_idx=i, + ln_scale=ln_scale, + dtg=dtg, + gated_attention=gated_attention, + value_residual=value_residual, + ) + for i in range(num_layers) + ] + ) + if rope_dims > 0: + head_dim = model_dim // num_heads + for block in self.blocks: + block.attn.rope_dims = rope_dims + block.attn.rotary = Rotary(head_dim, base=rope_base, train_seq_len=1024, rope_dims=rope_dims) + self.ve_layer_indices = [int(x) for x in ve_layers.split(",") if x.strip()] if ve_enabled else [] + kv_dim_ve = self._ve_target_dim + if self.ve_layer_indices: + self.ve_shared = ValueEmbedding(vocab_size, ve_dim, kv_dim_ve) + self.ve_layer_scales = nn.ParameterList( + [nn.Parameter(torch.ones(1, dtype=torch.float32)) for _ in self.ve_layer_indices] + ) + else: + self.ve_shared = None + self.ve_layer_scales = nn.ParameterList() + self.value_embeds = nn.ModuleList() # keep empty for compat + self.final_norm = RMSNorm() + self.lm_head = None if tie_embeddings else CastedLinear(model_dim, vocab_size, bias=False) + if self.lm_head is not None: + self.lm_head._zero_init = True + self.mtp_heads = nn.ModuleList( + [CastedLinear(model_dim, vocab_size, bias=False) for _ in range(mtp_num_heads)] + ) + for head in self.mtp_heads: + head._zero_init = True + if xsa_last_n > 0: + for i in range(max(0, num_layers - xsa_last_n), num_layers): + self.blocks[i].attn.use_xsa = True + self._init_weights() + def _init_weights(self) -> None: + if self.tie_embeddings: + nn.init.normal_(self.tok_emb.weight, mean=0.0, std=self.tied_embed_init_std) + n = self.num_layers + proj_scale = 1.0 / math.sqrt(2 * n) + # Init banks: orthogonal, with proj layers scaled down and out/down zero-init + for i in range(n): + nn.init.orthogonal_(self.qo_bank.data[i], gain=1.0) # Q + nn.init.zeros_(self.qo_bank.data[n + i]) # Out (zero init) + nn.init.orthogonal_(self.kv_bank.data[i], gain=1.0) # K + nn.init.orthogonal_(self.kv_bank.data[n + i], gain=1.0) # V + nn.init.orthogonal_(self.mlp_up_bank.data[i], gain=1.0) # MLP up + nn.init.zeros_(self.mlp_down_bank.data[i]) # MLP down (zero init) + # Scale proj layers (out_proj and mlp_down are "proj" layers) + self.qo_bank.data[n + i].mul_(proj_scale) + self.mlp_down_bank.data[i].mul_(proj_scale) + # Init remaining nn.Linear modules (bigram proj, mtp heads, lm_head) + for name, module in self.named_modules(): + if isinstance(module, nn.Linear): + if getattr(module, "_zero_init", False): + nn.init.zeros_(module.weight) + elif module.weight.ndim == 2 and module.weight.shape[0] >= 64 and module.weight.shape[1] >= 64: + nn.init.orthogonal_(module.weight, gain=1.0) + def _get_ve(self, layer_idx: int, input_ids: Tensor, ve_cache: dict | None = None) -> Tensor | None: + """Get value embedding for a specific layer using shared table + per-layer scale.""" + if self.ve_shared is None or layer_idx not in self.ve_layer_indices: + return None + if ve_cache is not None and 've' not in ve_cache: + ve_cache['ve'] = self.ve_shared(input_ids) + ve_base = ve_cache['ve'] if ve_cache is not None else self.ve_shared(input_ids) + ve_idx = self.ve_layer_indices.index(layer_idx) + return ve_base * self.ve_layer_scales[ve_idx].to(dtype=ve_base.dtype) + def forward(self, input_ids: Tensor, target_ids: Tensor) -> Tensor: + n = self.num_layers + x = self.tok_emb(input_ids) + if self.bigram is not None: + x = x + self.bigram(input_ids) + x = F.rms_norm(x, (x.size(-1),)) + x = self.smear(x) + x0 = x + v0 = None + skips: list[Tensor] = [] + ve_cache: dict = {} + for i in range(self.num_encoder_layers): + ve = self._get_ve(i, input_ids, ve_cache) + x, raw_v = self.blocks[i](x, x0, + self.qo_bank[i], self.kv_bank[i], self.kv_bank[n + i], + self.qo_bank[n + i], self.mlp_up_bank[i], self.mlp_down_bank[i], + v_embed=ve, v0=v0) + if v0 is None and raw_v is not None: + v0 = raw_v + skips.append(x) + for i in range(self.num_decoder_layers): + bi = self.num_encoder_layers + i + if skips: + scaled_skip = self.skip_weights[i].to(dtype=x.dtype)[None, None, :] * skips.pop() + if self.skip_gates is not None: + g = torch.sigmoid(self.skip_gates[i].to(dtype=x.dtype))[None, None, :] + x = torch.lerp(scaled_skip, x, g) + else: + x = x + scaled_skip + ve = self._get_ve(bi, input_ids, ve_cache) + x, _ = self.blocks[bi](x, x0, + self.qo_bank[bi], self.kv_bank[bi], self.kv_bank[n + bi], + self.qo_bank[n + bi], self.mlp_up_bank[bi], self.mlp_down_bank[bi], + v_embed=ve, v0=v0) + x = self.final_norm(x) + x_flat = x.reshape(-1, x.size(-1)) + targets = target_ids.reshape(-1) + if self.tie_embeddings: + logits_proj = F.linear(x_flat, self.tok_emb.weight) + else: + if self.lm_head is None: + raise RuntimeError("lm_head is required when tie_embeddings=False") + logits_proj = self.lm_head(x_flat) + logits = self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + main_loss = F.cross_entropy(logits.float(), targets, reduction="mean") + if self.training and self.mtp_num_heads > 0 and self.mtp_loss_weight > 0.0: + _, seqlen, dim = x.shape + mtp_loss_sum = x.new_zeros(()) + mtp_loss_count = 0 + for k, mtp_head in enumerate(self.mtp_heads): + valid_t = seqlen - (k + 1) + if valid_t <= 0: + continue + mtp_hidden = x[:, :valid_t, :].reshape(-1, dim) + mtp_targets = target_ids[:, k + 1 :].reshape(-1) + mtp_logits_proj = mtp_head(mtp_hidden) + mtp_logits = self.logit_softcap * torch.tanh(mtp_logits_proj / self.logit_softcap) + mtp_loss_sum = mtp_loss_sum + F.cross_entropy(mtp_logits.float(), mtp_targets, reduction="mean") + mtp_loss_count += 1 + if mtp_loss_count > 0: + main_loss = main_loss + self.mtp_loss_weight * (mtp_loss_sum / mtp_loss_count) + return main_loss + def forward_hidden(self, input_ids: Tensor) -> Tensor: + """Return final hidden states before lm_head.""" + n = self.num_layers + x = self.tok_emb(input_ids) + if self.bigram is not None: + x = x + self.bigram(input_ids) + x = F.rms_norm(x, (x.size(-1),)) + x = self.smear(x) + x0 = x + v0 = None + skips: list[Tensor] = [] + ve_cache: dict = {} + for i in range(self.num_encoder_layers): + ve = self._get_ve(i, input_ids, ve_cache) + x, raw_v = self.blocks[i](x, x0, + self.qo_bank[i], self.kv_bank[i], self.kv_bank[n + i], + self.qo_bank[n + i], self.mlp_up_bank[i], self.mlp_down_bank[i], + v_embed=ve, v0=v0) + if v0 is None and raw_v is not None: + v0 = raw_v + skips.append(x) + for i in range(self.num_decoder_layers): + bi = self.num_encoder_layers + i + if skips: + scaled_skip = self.skip_weights[i].to(dtype=x.dtype)[None, None, :] * skips.pop() + if self.skip_gates is not None: + g = torch.sigmoid(self.skip_gates[i].to(dtype=x.dtype))[None, None, :] + x = torch.lerp(scaled_skip, x, g) + else: + x = x + scaled_skip + ve = self._get_ve(bi, input_ids, ve_cache) + x, _ = self.blocks[bi](x, x0, + self.qo_bank[bi], self.kv_bank[bi], self.kv_bank[n + bi], + self.qo_bank[n + bi], self.mlp_up_bank[bi], self.mlp_down_bank[bi], + v_embed=ve, v0=v0) + return self.final_norm(x) + + def compute_logits(self, hidden_states: Tensor) -> Tensor: + """Compute logits from hidden states.""" + if self.tie_embeddings: + logits_proj = F.linear(hidden_states, self.tok_emb.weight) + else: + logits_proj = self.lm_head(hidden_states) + return self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + + def forward_logits(self, input_ids: Tensor) -> Tensor: + """Return logits (bsz, seq_len, vocab) without computing loss.""" + return self.compute_logits(self.forward_hidden(input_ids)) + +# --- Sliding window evaluation --- + +def eval_val_sliding( + args: Hyperparameters, + base_model: nn.Module, + rank: int, + world_size: int, + device: torch.device, + val_tokens: Tensor, + base_bytes_lut: Tensor, + has_leading_space_lut: Tensor, + is_boundary_token_lut: Tensor, + stride: int, + batch_seqs: int = 32, + eval_seq_len: int | None = None, +) -> tuple[float, float]: + """Sliding window evaluation: each token scored with maximum context.""" + seq_len = eval_seq_len or args.train_seq_len + total_tokens = val_tokens.numel() - 1 + window_starts = [ws for ws in range(0, total_tokens, stride) + if min(ws + seq_len, total_tokens) - ws >= 1] + total_windows = len(window_starts) + my_s = (total_windows * rank) // world_size + my_e = (total_windows * (rank + 1)) // world_size + my_windows = window_starts[my_s:my_e] + loss_sum = torch.zeros((), device=device, dtype=torch.float64) + token_count = torch.zeros((), device=device, dtype=torch.float64) + byte_count = torch.zeros((), device=device, dtype=torch.float64) + base_model.eval() + compiled_logits = torch.compile(base_model.forward_logits, dynamic=False, fullgraph=True) + with torch.inference_mode(): + for bi in range(0, len(my_windows), batch_seqs): + batch_ws = my_windows[bi:bi + batch_seqs] + bsz = len(batch_ws) + x_batch = torch.zeros(bsz, seq_len, dtype=torch.int64, device=device) + y_batch = torch.zeros(bsz, seq_len, dtype=torch.int64, device=device) + wlens: list[int] = [] + for i, ws in enumerate(batch_ws): + end = min(ws + seq_len, total_tokens) + wlen = end - ws + wlens.append(wlen) + chunk = val_tokens[ws:end + 1].to(dtype=torch.int64, device=device) + x_batch[i, :wlen] = chunk[:-1] + y_batch[i, :wlen] = chunk[1:] + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + logits = compiled_logits(x_batch) + nll = F.cross_entropy( + logits.reshape(-1, logits.size(-1)).float(), + y_batch.reshape(-1), + reduction="none", + ).reshape(bsz, seq_len) + for i, ws in enumerate(batch_ws): + wlen = wlens[i] + s = 0 if ws == 0 else max(wlen - stride, 0) + scored_nll = nll[i, s:wlen].to(torch.float64) + loss_sum += scored_nll.sum() + token_count += float(wlen - s) + tgt = y_batch[i, s:wlen] + prev = x_batch[i, s:wlen] + tb = base_bytes_lut[tgt].to(torch.float64) + tb += (has_leading_space_lut[tgt] & ~is_boundary_token_lut[prev]).to(torch.float64) + byte_count += tb.sum() + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(token_count, op=dist.ReduceOp.SUM) + dist.all_reduce(byte_count, op=dist.ReduceOp.SUM) + val_loss = (loss_sum / token_count).item() + bits_per_token = val_loss / math.log(2.0) + tokens_per_byte = token_count.item() / byte_count.item() + base_model.train() + return val_loss, bits_per_token * tokens_per_byte + + +def ttt_adapt_adamw( + args, base_model: nn.Module, device: torch.device, + val_tokens: Tensor, rank: int = 0, world_size: int = 1, log0=print, +) -> None: + """AdamW TTT: fine-tune on val data BEFORE quantization (PR #1006 style).""" + seq_len = args.train_seq_len + total_seqs = (val_tokens.numel() - 1) // seq_len + batch_seqs = args.ttt_batch_seqs + if args.ttt_freeze_blocks > 0: + for i, block in enumerate(base_model.blocks): + if i < args.ttt_freeze_blocks: + for p in block.parameters(): + p.requires_grad_(False) + ttt_params = [p for p in base_model.parameters() if p.requires_grad] + log0(f"ttt_adamw:params trainable={sum(p.numel() for p in ttt_params)} " + f"frozen={sum(p.numel() for p in base_model.parameters() if not p.requires_grad)}") + optimizer = torch.optim.AdamW(ttt_params, lr=args.ttt_lr, weight_decay=0.0) + scheduler = None + if args.ttt_cosine_decay: + scheduler = torch.optim.lr_scheduler.CosineAnnealingLR( + optimizer, T_max=args.ttt_epochs, eta_min=args.ttt_lr * 0.1) + my_start = (total_seqs * rank) // world_size + my_end = (total_seqs * (rank + 1)) // world_size + base_model.train() + t0 = time.perf_counter() + for epoch in range(args.ttt_epochs): + epoch_loss_sum = torch.zeros((), device=device, dtype=torch.float64) + epoch_tokens = torch.zeros((), device=device, dtype=torch.float64) + for bs in range(my_start, my_end, batch_seqs): + be = min(bs + batch_seqs, my_end) + raw_start = bs * seq_len + raw_end = be * seq_len + 1 + if raw_end > val_tokens.numel(): + continue + local = val_tokens[raw_start:raw_end].to(device=device, dtype=torch.int64) + x = local[:-1].reshape(-1, seq_len) + y = local[1:].reshape(-1, seq_len) + optimizer.zero_grad(set_to_none=True) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + loss = base_model(x, y) + loss.backward() + if world_size > 1: + for p in ttt_params: + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG) + torch.nn.utils.clip_grad_norm_(ttt_params, args.ttt_grad_clip) + optimizer.step() + epoch_loss_sum += loss.detach().to(torch.float64) * float(y.numel()) + epoch_tokens += float(y.numel()) + if world_size > 1: + dist.all_reduce(epoch_loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(epoch_tokens, op=dist.ReduceOp.SUM) + epoch_avg_loss = epoch_loss_sum.item() / max(epoch_tokens.item(), 1) + if scheduler is not None: + scheduler.step() + log0(f"ttt_adamw:epoch {epoch+1}/{args.ttt_epochs} loss:{epoch_avg_loss:.4f} " + f"time:{time.perf_counter() - t0:.1f}s") + for p in base_model.parameters(): + p.requires_grad_(True) + base_model.eval() + log0(f"ttt_adamw:done elapsed={time.perf_counter() - t0:.1f}s") + + +def generate_autoregressive_calib(model, device, num_seqs=64, seq_len=2048, + vocab_size=1024, temperature=0.8, batch_size=8, seed=42): + """Generate sequences autoregressively from the model for GPTQ calibration. + No external data accessed — fully self-contained.""" + model.eval() + rng = torch.Generator(device=device) + rng.manual_seed(seed) + all_tokens = [] + with torch.inference_mode(), torch.autocast(device_type="cuda", dtype=torch.bfloat16): + for batch_start in range(0, num_seqs, batch_size): + bs = min(batch_size, num_seqs - batch_start) + tokens = torch.randint(0, vocab_size, (bs, 1), device=device, generator=rng) + for pos in range(seq_len - 1): + logits = model.forward_logits(tokens) + next_logit = logits[:, -1, :] + probs = torch.softmax(next_logit / temperature, dim=-1) + next_tok = torch.multinomial(probs, 1, generator=rng) + tokens = torch.cat([tokens, next_tok], dim=1) + for i in range(bs): + all_tokens.append(tokens[i:i+1]) + return all_tokens + + +def collect_hessians_from_tokens(hessian_model, token_seqs, device): + """Collect H = X^T X from pre-generated token sequences.""" + hessians = {} + hooks = [] + for name, module in hessian_model.named_modules(): + if isinstance(module, CastedLinear): + param_name = name + ".weight" + cols = module.weight.shape[1] + hessians[param_name] = torch.zeros(cols, cols, dtype=torch.float32, device='cpu') + h = module.register_forward_hook(_make_input_hook(hessians, param_name)) + hooks.append(h) + hessian_model.eval() + with torch.inference_mode(), torch.autocast(device_type="cuda", dtype=torch.bfloat16): + for seq in token_seqs: + x = seq[:, :-1].to(device) + y = seq[:, 1:].to(device) + hessian_model(x, y) + for h in hooks: + h.remove() + num_batches = len(token_seqs) + for name in hessians: + H = hessians[name] + H /= num_batches + damp = 0.01 * torch.diag(H).mean().clamp_min(1e-6) + H += damp * torch.eye(H.shape[0]) + hessians[name] = H + return hessians + + +# --- GPTQ-lite int6 quantization --- + +def _classify_param(name: str) -> str: + if "tok_emb" in name or "lm_head" in name: + return "embed" + if ".mlp." in name: + return "mlp" + if ".attn." in name or (".proj." in name and ".mlp." not in name): + return "attn" + return "other" +def sdclip_scale(t32: Tensor, clip_sigmas: float, clip_range: int) -> Tensor: + """SDClip: scale = clip_sigmas * std(row) / clip_range (per-row for 2D, global for 1D).""" + if t32.ndim == 2: + row_std = t32.std(dim=1).clamp_min(1e-10) + return (clip_sigmas * row_std / clip_range).clamp_min(1.0 / clip_range).to(torch.float16) + amax = t32.abs().max().item() + return torch.tensor(amax / clip_range if amax > 0 else 1.0, dtype=torch.float16) + +def quantize_int6_per_row(t: Tensor, clip_range: int = 31, clip_sigmas: float = 12.85) -> tuple[Tensor, Tensor]: + t32 = t.float() + s = sdclip_scale(t32, clip_sigmas, clip_range) + if t32.ndim == 2: + q = torch.clamp(torch.round(t32 / s.float()[:, None]), -clip_range, clip_range).to(torch.int8) + else: + q = torch.clamp(torch.round(t32 / s.float()), -clip_range, clip_range).to(torch.int8) + return q, s + +def quantize_int6_gptq(weight, hessian=None, clip_range=31, block_size=128, clip_sigmas=12.85): + """Full GPTQ: Hessian-aware int6 quantization with Cholesky error compensation. + Uses SDClip (k * std per row) for scale computation. + If hessian is None, falls back to SDClip-only quantization.""" + t32 = weight.float() + if t32.ndim != 2 or hessian is None: + return quantize_int6_per_row(t32, clip_range=clip_range, clip_sigmas=clip_sigmas) + rows, cols = t32.shape + H = hessian.float().clone() + dead = torch.diag(H) == 0 + H[dead, dead] = 1 + damp = 0.01 * torch.mean(torch.diag(H)) + H[torch.arange(cols), torch.arange(cols)] += damp + perm = torch.argsort(torch.diag(H), descending=True) + inv_perm = torch.argsort(perm) + W = t32[:, perm].clone() + W[:, dead[perm]] = 0 + H = H[perm][:, perm] + Hinv = torch.linalg.cholesky(H) + Hinv = torch.cholesky_inverse(Hinv) + Hinv = torch.linalg.cholesky(Hinv, upper=True) + # SDClip: single-pass scale from k * std(row) + s = sdclip_scale(t32, clip_sigmas, clip_range) + sf = s.float() + Q = torch.zeros_like(W, dtype=torch.int8) + W_work = W.clone() + for i1 in range(0, cols, block_size): + i2 = min(i1 + block_size, cols) + count = i2 - i1 + W1 = W_work[:, i1:i2].clone() + Q1 = torch.zeros(rows, count, dtype=torch.int8) + Err1 = torch.zeros(rows, count) + Hinv1 = Hinv[i1:i2, i1:i2] + for i in range(count): + w = W1[:, i] + d = Hinv1[i, i] + q = torch.clamp(torch.round(w / sf), -clip_range, clip_range).to(torch.int8) + Q1[:, i] = q + err = (w - q.float() * sf) / d + W1[:, i:] -= err.unsqueeze(1) * Hinv1[i, i:].unsqueeze(0) + Err1[:, i] = err + Q[:, i1:i2] = Q1 + if i2 < cols: + W_work[:, i2:] -= Err1 @ Hinv[i1:i2, i2:] + Q = Q[:, inv_perm] + return Q, s + +def _quantize_int6_percentile(t32, clip_range=31, clip_sigmas=12.85): + """Fallback: SDClip-based (kept for API compat, now uses SDClip not percentile).""" + return quantize_int6_per_row(t32, clip_range=clip_range, clip_sigmas=clip_sigmas) + +def _unbank_state_dict(sd: dict[str, Tensor], num_layers: int) -> dict[str, Tensor]: + """Convert 3D bank tensors into individual 2D tensors with standard names.""" + out: dict[str, Tensor] = {} + n = num_layers + for name, tensor in sd.items(): + if name == "qo_bank": + for i in range(n): + out[f"blocks.{i}.attn.c_q.weight"] = tensor[i] + out[f"blocks.{i}.attn.proj.weight"] = tensor[n + i] + elif name == "kv_bank": + for i in range(n): + out[f"blocks.{i}.attn.c_k.weight"] = tensor[i] + out[f"blocks.{i}.attn.c_v.weight"] = tensor[n + i] + elif name == "mlp_up_bank": + for i in range(n): + out[f"blocks.{i}.mlp.fc.weight"] = tensor[i] + elif name == "mlp_down_bank": + for i in range(n): + out[f"blocks.{i}.mlp.proj.weight"] = tensor[i] + else: + out[name] = tensor + return out + +def _rebank_state_dict(sd: dict[str, Tensor], num_layers: int, template_sd: dict[str, Tensor]) -> dict[str, Tensor]: + """Convert individual 2D tensors back into 3D bank tensors.""" + out: dict[str, Tensor] = {} + n = num_layers + # Reconstruct banks from individual weight keys + qo_slices = [None] * (2 * n) + kv_slices = [None] * (2 * n) + up_slices = [None] * n + down_slices = [None] * n + consumed = set() + for i in range(n): + qk = f"blocks.{i}.attn.c_q.weight" + if qk in sd: + qo_slices[i] = sd[qk] + consumed.add(qk) + ok = f"blocks.{i}.attn.proj.weight" + if ok in sd: + qo_slices[n + i] = sd[ok] + consumed.add(ok) + kk = f"blocks.{i}.attn.c_k.weight" + if kk in sd: + kv_slices[i] = sd[kk] + consumed.add(kk) + vk = f"blocks.{i}.attn.c_v.weight" + if vk in sd: + kv_slices[n + i] = sd[vk] + consumed.add(vk) + fk = f"blocks.{i}.mlp.fc.weight" + if fk in sd: + up_slices[i] = sd[fk] + consumed.add(fk) + dk = f"blocks.{i}.mlp.proj.weight" + if dk in sd: + down_slices[i] = sd[dk] + consumed.add(dk) + out["qo_bank"] = torch.stack(qo_slices).to(dtype=template_sd["qo_bank"].dtype) + out["kv_bank"] = torch.stack(kv_slices).to(dtype=template_sd["kv_bank"].dtype) + out["mlp_up_bank"] = torch.stack(up_slices).to(dtype=template_sd["mlp_up_bank"].dtype) + out["mlp_down_bank"] = torch.stack(down_slices).to(dtype=template_sd["mlp_down_bank"].dtype) + for name, tensor in sd.items(): + if name not in consumed: + out[name] = tensor + return out + +# --- Non-banked model for Hessian collection --- +# This mirrors the unbanked state dict keys: blocks.{i}.attn.c_q/c_k/c_v/proj, blocks.{i}.mlp.fc/proj + +class _HessianAttn(nn.Module): + """Non-banked attention with CastedLinear layers for Hessian hooks.""" + def __init__(self, dim, num_heads, num_kv_heads, rope_base, qk_gain_init): + super().__init__() + self.num_heads, self.num_kv_heads = num_heads, num_kv_heads + self.head_dim = dim // num_heads + kv_dim = num_kv_heads * self.head_dim + self.c_q = CastedLinear(dim, dim, bias=False) + self.c_k = CastedLinear(dim, kv_dim, bias=False) + self.c_v = CastedLinear(dim, kv_dim, bias=False) + self.proj = CastedLinear(dim, dim, bias=False) + self.q_gain = nn.Parameter(torch.full((num_heads,), qk_gain_init, dtype=torch.float32)) + self.rope_dims = 0 + self.rotary = Rotary(self.head_dim, base=rope_base, train_seq_len=1024) + self.use_xsa = False + def _xsa_efficient(self, y, v): + B, T, H, D = y.shape; Hkv = v.size(-2); group = H // Hkv + y_g = y.reshape(B, T, Hkv, group, D) + vn = F.normalize(v, dim=-1).unsqueeze(-2) + proj = (y_g * vn).sum(dim=-1, keepdim=True) * vn + return (y_g - proj).reshape(B, T, H, D) + def forward(self, x, v_embed=None): + bsz, seqlen, dim = x.shape + q = self.c_q(x).reshape(bsz, seqlen, self.num_heads, self.head_dim) + k = self.c_k(x).reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + v = self.c_v(x) + if v_embed is not None: + v = v + v_embed + v = v.reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + q = F.rms_norm(q, (q.size(-1),)) + k = F.rms_norm(k, (k.size(-1),)) + cos, sin = self.rotary(seqlen, x.device, q.dtype) + q = apply_rotary_emb(q, cos, sin, self.rope_dims) + k = apply_rotary_emb(k, cos, sin, self.rope_dims) + q = q * self.q_gain.to(dtype=q.dtype)[None, None, :, None] + if HAS_FA3: + y = flash_attn_3_func(q, k, v, causal=True) + else: + q2, k2, v2 = q.transpose(1, 2), k.transpose(1, 2), v.transpose(1, 2) + if k2.size(1) != q2.size(1): + reps = q2.size(1) // k2.size(1) + k2 = k2.repeat_interleave(reps, dim=1) + v2 = v2.repeat_interleave(reps, dim=1) + y = F.scaled_dot_product_attention(q2, k2, v2, is_causal=True).transpose(1, 2) + if self.use_xsa: + y = self._xsa_efficient(y, v) + return self.proj(y.reshape(bsz, seqlen, dim)) + +class _HessianMLP(nn.Module): + """Non-banked MLP with CastedLinear layers for Hessian hooks.""" + def __init__(self, dim, mlp_mult): + super().__init__() + self.fc = CastedLinear(dim, int(mlp_mult * dim), bias=False) + self.proj = CastedLinear(int(mlp_mult * dim), dim, bias=False) + def forward(self, x): + return self.proj(F.leaky_relu(self.fc(x), negative_slope=0.5).square()) + +class _HessianBlock(nn.Module): + def __init__(self, dim, num_heads, num_kv_heads, mlp_mult, rope_base, qk_gain_init, layer_idx=0, ln_scale=False): + super().__init__() + self.attn_norm = RMSNorm() + self.mlp_norm = RMSNorm() + self.attn = _HessianAttn(dim, num_heads, num_kv_heads, rope_base, qk_gain_init) + self.mlp = _HessianMLP(dim, mlp_mult) + self.attn_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.mlp_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.resid_mix = nn.Parameter(torch.stack((torch.ones(dim), torch.zeros(dim))).float()) + self.ln_scale_factor = 1.0 / math.sqrt(layer_idx + 1) if ln_scale else 1.0 + def forward(self, x, x0, v_embed=None): + mix = self.resid_mix.to(dtype=x.dtype) + x_in = mix[0][None, None, :] * x + mix[1][None, None, :] * x0 + attn_out = self.attn(self.attn_norm(x_in) * self.ln_scale_factor, v_embed=v_embed) + x_out = x_in + self.attn_scale.to(dtype=x_in.dtype)[None, None, :] * attn_out + x_out = x_out + self.mlp_scale.to(dtype=x_out.dtype)[None, None, :] * self.mlp(self.mlp_norm(x_out) * self.ln_scale_factor) + return x_out + +class _HessianGPT(nn.Module): + """Non-banked GPT model matching unbanked state dict keys for Hessian collection.""" + def __init__(self, vocab_size, num_layers, model_dim, num_heads, num_kv_heads, + mlp_mult, tie_embeddings, logit_softcap, rope_base, qk_gain_init, + bigram_vocab_size=0, bigram_dim=128, xsa_last_n=0, + rope_dims=0, ln_scale=False, + ve_enabled=False, ve_dim=128, ve_layers="9,10"): + super().__init__() + self.tie_embeddings = tie_embeddings + self.logit_softcap = logit_softcap + self.num_layers = num_layers + self.tok_emb = nn.Embedding(vocab_size, model_dim) + self.bigram = BigramHashEmbedding(bigram_vocab_size, bigram_dim, model_dim, trigram=bool(int(os.environ.get("TRIGRAM", "0")))) if bigram_vocab_size > 0 else None + self.smear = SmearGate(model_dim) + self.num_encoder_layers = num_layers // 2 + self.num_decoder_layers = num_layers - self.num_encoder_layers + self.num_skip_weights = min(self.num_encoder_layers, self.num_decoder_layers) + self.skip_weights = nn.Parameter(torch.ones(self.num_skip_weights, model_dim, dtype=torch.float32)) + self.blocks = nn.ModuleList([ + _HessianBlock(model_dim, num_heads, num_kv_heads, mlp_mult, rope_base, qk_gain_init, + layer_idx=i, ln_scale=ln_scale) + for i in range(num_layers) + ]) + if rope_dims > 0: + head_dim = model_dim // num_heads + for block in self.blocks: + block.attn.rope_dims = rope_dims + block.attn.rotary = Rotary(head_dim, base=rope_base, train_seq_len=1024, rope_dims=rope_dims) + if xsa_last_n > 0: + for i in range(max(0, num_layers - xsa_last_n), num_layers): + self.blocks[i].attn.use_xsa = True + kv_dim = num_kv_heads * (model_dim // num_heads) + self.ve_layer_indices = [int(x) for x in ve_layers.split(",") if x.strip()] if ve_enabled else [] + if self.ve_layer_indices: + self.ve_shared = ValueEmbedding(vocab_size, ve_dim, kv_dim) + self.ve_layer_scales = nn.ParameterList([nn.Parameter(torch.ones(1, dtype=torch.float32)) for _ in self.ve_layer_indices]) + else: + self.ve_shared = None + self.ve_layer_scales = nn.ParameterList() + self.final_norm = RMSNorm() + self.lm_head = None if tie_embeddings else CastedLinear(model_dim, vocab_size, bias=False) + def _get_ve(self, layer_idx, input_ids, ve_cache): + if self.ve_shared is None or layer_idx not in self.ve_layer_indices: + return None + if 've' not in ve_cache: + ve_cache['ve'] = self.ve_shared(input_ids) + ve_idx = self.ve_layer_indices.index(layer_idx) + return ve_cache['ve'] * self.ve_layer_scales[ve_idx].to(dtype=ve_cache['ve'].dtype) + def forward(self, input_ids, target_ids): + x = self.tok_emb(input_ids) + if self.bigram is not None: + x = x + self.bigram(input_ids) + x = F.rms_norm(x, (x.size(-1),)) + x = self.smear(x) + x0 = x + skips = [] + ve_cache = {} + for i in range(self.num_encoder_layers): + ve = self._get_ve(i, input_ids, ve_cache) + x = self.blocks[i](x, x0, v_embed=ve) + skips.append(x) + for i in range(self.num_decoder_layers): + bi = self.num_encoder_layers + i + if skips: + x = x + self.skip_weights[i].to(dtype=x.dtype)[None, None, :] * skips.pop() + ve = self._get_ve(bi, input_ids, ve_cache) + x = self.blocks[bi](x, x0, v_embed=ve) + x = self.final_norm(x) + x_flat = x.reshape(-1, x.size(-1)) + targets = target_ids.reshape(-1) + logits_proj = F.linear(x_flat, self.tok_emb.weight) if self.tie_embeddings else self.lm_head(x_flat) + logits = self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + return F.cross_entropy(logits.float(), targets, reduction="mean") + +def _make_input_hook(hessians, pname): + def hook_fn(module, input, output): + x = input[0].detach().float() + if x.ndim == 3: + x = x.reshape(-1, x.shape[-1]) + hessians[pname] += (x.T @ x).cpu() + return hook_fn + +def _make_output_hook(hessians, pname): + def hook_fn(module, input, output): + x = output.detach().float() + if x.ndim == 3: + x = x.reshape(-1, x.shape[-1]) + hessians[pname] += (x.T @ x).cpu() + return hook_fn + +def collect_hessians(hessian_model, train_loader, args, device, grad_accum_steps, num_batches=256): + """Run calibration batches through a non-banked model, collecting H = X^T X for each CastedLinear. + Also captures the pre-logit hidden states for tok_emb.weight GPTQ (embedding GPTQ).""" + hessians = {} + hooks = [] + for name, module in hessian_model.named_modules(): + if isinstance(module, CastedLinear): + param_name = name + ".weight" + cols = module.weight.shape[1] + hessians[param_name] = torch.zeros(cols, cols, dtype=torch.float32, device='cpu') + h = module.register_forward_hook(_make_input_hook(hessians, param_name)) + hooks.append(h) + # Embedding GPTQ: capture Hessian for tok_emb.weight via pre-logit hidden states. + # The tied embedding acts as a linear layer: logits = hidden @ tok_emb.weight.T + # so the activation Hessian is X^T X where X = final_norm output. + embed_hessian_key = "tok_emb.weight" + embed_dim = hessian_model.tok_emb.weight.shape[1] + hessians[embed_hessian_key] = torch.zeros(embed_dim, embed_dim, dtype=torch.float32, device='cpu') + embed_hook = hessian_model.final_norm.register_forward_hook( + _make_output_hook(hessians, embed_hessian_key) + ) + hooks.append(embed_hook) + hessian_model.eval() + with torch.inference_mode(), torch.autocast(device_type="cuda", dtype=torch.bfloat16): + for _ in range(num_batches): + x, y = train_loader.next_batch(args.train_batch_tokens, args.train_seq_len, grad_accum_steps) + hessian_model(x, y) + for h in hooks: + h.remove() + for name in hessians: + H = hessians[name] + H /= num_batches + damp = 0.01 * torch.diag(H).mean().clamp_min(1e-6) + H += damp * torch.eye(H.shape[0]) + hessians[name] = H + hessian_model.train() + return hessians + +def mixed_quantize_int6(state_dict: dict[str, Tensor], int6_cats: set[str], hessians: dict[str, Tensor] | None = None, + matrix_bits: int = 6, embed_bits: int = 8, + matrix_clip_sigmas: float = 12.85, embed_clip_sigmas: float = 20.0): + """Mixed-precision GPTQ quantization. + - mlp/attn weights: int6 with SDClip (k=matrix_clip_sigmas) and Hessian compensation + - tok_emb/lm_head: GPTQ with int8 and SDClip (k=embed_clip_sigmas) + - small tensors: passthrough float16 + """ + num_layers_total = max( + (int(k.split(".")[1]) for k in state_dict if k.startswith("blocks.")), + default=0, + ) + 1 + result: dict[str, Tensor] = {} + meta: dict[str, object] = {} + for name, tensor in state_dict.items(): + t = tensor.detach().cpu().contiguous() + cat = _classify_param(name) + if not t.is_floating_point() or t.numel() <= 65536: + result[name] = t.to(torch.float16) if t.is_floating_point() else t + meta[name] = "passthrough" + continue + if any(p in name for p in CONTROL_TENSOR_NAME_PATTERNS): + result[name] = t.float() + meta[name] = "passthrough_ctrl" + continue + if cat == "embed" and t.ndim >= 1: + # GPTQ on embeddings with int8 + SDClip + cr = 2 ** (embed_bits - 1) - 1 # 127 for int8 + H = hessians.get(name) if hessians else None + if H is not None: + q, s = quantize_int6_gptq(t, hessian=H, clip_range=cr, clip_sigmas=embed_clip_sigmas) + else: + q, s = quantize_int6_per_row(t, clip_range=cr, clip_sigmas=embed_clip_sigmas) + result[name + ".q"] = q + result[name + ".scale"] = s + meta[name] = {"type": f"int{embed_bits}"} + elif cat in int6_cats and t.ndim >= 1: + cr = 2 ** (matrix_bits - 1) - 1 # 31 for int6 + H = hessians.get(name) if hessians else None + if H is not None: + q, s = quantize_int6_gptq(t, hessian=H, clip_range=cr, clip_sigmas=matrix_clip_sigmas) + else: + q, s = quantize_int6_per_row(t, clip_range=cr, clip_sigmas=matrix_clip_sigmas) + result[name + ".q"] = q + result[name + ".scale"] = s + meta[name] = {"type": f"int{matrix_bits}"} + else: + q, s = quantize_float_tensor(t) + result[name + ".q"] = q + result[name + ".scale"] = s + meta[name] = {"type": "int8"} + return result, meta +def dequantize_mixed_int6(result: dict[str, Tensor], meta: dict[str, object], + template_sd: dict[str, Tensor]) -> dict[str, Tensor]: + out: dict[str, Tensor] = {} + for name, orig in template_sd.items(): + info = meta.get(name) + if info is None: + continue + orig_dtype = orig.dtype + if info in ("passthrough", "passthrough_ctrl", "passthrough_fp16"): + t = result[name] + if t.dtype == torch.float16 and orig_dtype in (torch.float32, torch.bfloat16): + t = t.to(orig_dtype) + out[name] = t + continue + q, s = result[name + ".q"], result[name + ".scale"] + if s.ndim > 0: + out[name] = (q.float() * s.float().view(q.shape[0], *([1] * (q.ndim - 1)))).to(orig_dtype) + else: + out[name] = (q.float() * float(s.item())).to(orig_dtype) + return out + +# --- Byte-shuffle + Brotli compression --- + +_BSHF_MAGIC = b'BSHF' + +def _byte_shuffle(data: bytes, stride: int = 2) -> bytes: + """Interleave bytes by stride before compression (improves Brotli ratio on quantized weights).""" + if stride <= 1 or len(data) < stride: + return data + src = np.frombuffer(data, dtype=np.uint8) + n = len(src) + out = np.empty(n, dtype=np.uint8) + dest_off = 0 + for pos in range(stride): + chunk = src[pos::stride] + out[dest_off:dest_off + len(chunk)] = chunk + dest_off += len(chunk) + return _BSHF_MAGIC + bytes([stride]) + out.tobytes() + +def _byte_unshuffle(data: bytes) -> bytes: + """Reverse byte shuffle.""" + if len(data) < 5 or data[:4] != _BSHF_MAGIC: + return data + stride = data[4] + if stride < 2: + return data[5:] + payload = np.frombuffer(data, dtype=np.uint8, offset=5) + n = len(payload) + out = np.empty(n, dtype=np.uint8) + src_off = 0 + for pos in range(stride): + chunk_len = n // stride + (1 if pos < n % stride else 0) + out[pos::stride][:chunk_len] = payload[src_off:src_off + chunk_len] + src_off += chunk_len + return out.tobytes() + +def _compress_brotli(data: bytes) -> bytes: + """Byte-shuffle then Brotli compress.""" + import brotli + shuffled = _byte_shuffle(data, stride=2) + return brotli.compress(shuffled, quality=11) + +def _decompress_brotli(data: bytes) -> bytes: + """Brotli decompress then byte-unshuffle.""" + import brotli + raw = brotli.decompress(data) + return _byte_unshuffle(raw) + +# --- Training --- + +def main() -> None: + code = Path(__file__).read_text(encoding="utf-8") + args = Hyperparameters() + # zeropower_via_newtonschulz5 runs eagerly with bmm -- do NOT compile + distributed = "RANK" in os.environ and "WORLD_SIZE" in os.environ + rank = int(os.environ.get("RANK", "0")) + world_size = int(os.environ.get("WORLD_SIZE", "1")) + local_rank = int(os.environ.get("LOCAL_RANK", "0")) + if world_size <= 0: + raise ValueError(f"WORLD_SIZE must be positive, got {world_size}") + if 8 % world_size != 0: + raise ValueError(f"WORLD_SIZE={world_size} must divide 8 so grad_accum_steps stays integral") + grad_accum_steps = 8 // world_size + grad_scale = 1.0 / grad_accum_steps + if not torch.cuda.is_available(): + raise RuntimeError("CUDA is required") + device = torch.device("cuda", local_rank) + torch.cuda.set_device(device) + if distributed: + dist.init_process_group(backend="nccl", device_id=device) + dist.barrier() + master_process = rank == 0 + torch.backends.cuda.matmul.allow_tf32 = True + torch.backends.cudnn.allow_tf32 = True + from torch.backends.cuda import enable_cudnn_sdp, enable_flash_sdp, enable_math_sdp, enable_mem_efficient_sdp + enable_cudnn_sdp(False) + enable_flash_sdp(True) + enable_mem_efficient_sdp(False) + enable_math_sdp(False) + logfile = None + if master_process: + os.makedirs("logs", exist_ok=True) + logfile = f"logs/{args.run_id}.txt" + print(logfile) + def log0(msg: str, console: bool = True) -> None: + if not master_process: + return + if console: + print(msg) + if logfile is not None: + with open(logfile, "a", encoding="utf-8") as f: + print(msg, file=f) + log0(code, console=False) + log0("=" * 100, console=False) + log0(f"Running Python {sys.version}", console=False) + log0(f"Running PyTorch {torch.__version__}", console=False) + log0( + subprocess.run(["nvidia-smi"], stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, check=False).stdout, + console=False, + ) + log0("=" * 100, console=False) + random.seed(args.seed) + np.random.seed(args.seed) + torch.manual_seed(args.seed) + torch.cuda.manual_seed_all(args.seed) + if not args.tokenizer_path.endswith(".model"): + raise ValueError(f"Script only setup for SentencePiece .model file: {args.tokenizer_path}") + sp = spm.SentencePieceProcessor(model_file=args.tokenizer_path) + if int(sp.vocab_size()) != args.vocab_size: + raise ValueError( + f"VOCAB_SIZE={args.vocab_size} does not match tokenizer vocab_size={int(sp.vocab_size())}" + ) + dataset_dir = Path(args.data_path).resolve() + actual_train_files = len(list(dataset_dir.glob("fineweb_train_*.bin"))) + effective_eval_seq_len = args.eval_seq_len if args.eval_seq_len > 0 else args.train_seq_len + val_seq_len = max(args.train_seq_len, effective_eval_seq_len) + val_tokens = load_validation_tokens(args.val_files, val_seq_len) + base_bytes_lut, has_leading_space_lut, is_boundary_token_lut = build_sentencepiece_luts( + sp, args.vocab_size, device + ) + log0(f"val_bpb:enabled tokenizer_kind=sentencepiece tokenizer_path={args.tokenizer_path}") + log0(f"train_loader:dataset:{dataset_dir.name} train_shards:{actual_train_files}") + log0(f"val_loader:shards pattern={args.val_files} tokens:{val_tokens.numel() - 1}") + CastedLinear._qat_enabled = args.qat_enabled + base_model = GPT( + vocab_size=args.vocab_size, + num_layers=args.num_layers, + model_dim=args.model_dim, + num_heads=args.num_heads, + num_kv_heads=args.num_kv_heads, + mlp_mult=args.mlp_mult, + tie_embeddings=args.tie_embeddings, + tied_embed_init_std=args.tied_embed_init_std, + logit_softcap=args.logit_softcap, + rope_base=args.rope_base, + qk_gain_init=args.qk_gain_init, + mtp_num_heads=args.mtp_num_heads, + mtp_loss_weight=args.mtp_loss_weight, + bigram_vocab_size=args.bigram_vocab_size, + bigram_dim=args.bigram_dim, + xsa_last_n=args.xsa_last_n, + rope_dims=args.rope_dims, + ln_scale=args.ln_scale, + dtg=args.dtg_enabled, + ve_enabled=args.ve_enabled, + ve_dim=args.ve_dim, + ve_layers=args.ve_layers, + gated_attention=args.gated_attention, + value_residual=args.value_residual, + skip_gates_enabled=args.skip_gates_enabled, + ).to(device).bfloat16() + # Banks stay FP32 (like CastedLinear weights), cast to BF16 in forward + base_model.qo_bank.data = base_model.qo_bank.data.float() + base_model.kv_bank.data = base_model.kv_bank.data.float() + base_model.mlp_up_bank.data = base_model.mlp_up_bank.data.float() + base_model.mlp_down_bank.data = base_model.mlp_down_bank.data.float() + for module in base_model.modules(): + if isinstance(module, CastedLinear): + module.float() + restore_low_dim_params_to_fp32(base_model) + # No DDP -- Parallel Muon handles bank grad communication via reduce-scatter, + # and non-bank grads are manually all-reduced before Adam steps. + compiled_model = torch.compile(base_model, dynamic=False, fullgraph=True) + model = compiled_model + + # Optimizer split: + # - 4 parameter banks -> Muon (batched Newton-Schulz) + # - token embedding -> Adam + # - scalars/control tensors -> Adam + # - bigram proj, mtp heads, VE proj -> Adam (small matrix params not worth banking) + matrix_params = [ + base_model.qo_bank, base_model.kv_bank, + base_model.mlp_up_bank, base_model.mlp_down_bank, + ] + block_named_params = list(base_model.blocks.named_parameters()) + scalar_params = [ + p + for name, p in block_named_params + if p.ndim < 2 or any(pattern in name for pattern in CONTROL_TENSOR_NAME_PATTERNS) + ] + if base_model.skip_weights.numel() > 0: + scalar_params.append(base_model.skip_weights) + if base_model.skip_gates is not None and base_model.skip_gates.numel() > 0: + scalar_params.append(base_model.skip_gates) + scalar_params.append(base_model.smear.gate) + if base_model.bigram is not None: + scalar_params.append(base_model.bigram.scale) + token_lr = args.tied_embed_lr if args.tie_embeddings else args.embed_lr + tok_params = [{"params": [base_model.tok_emb.weight], "lr": token_lr, "base_lr": token_lr}] + if base_model.bigram is not None: + tok_params.append({"params": [base_model.bigram.embed.weight], "lr": token_lr, "base_lr": token_lr}) + if base_model.bigram.proj is not None: + scalar_params.append(base_model.bigram.proj.weight) + if base_model.ve_shared is not None: + tok_params.append({"params": [base_model.ve_shared.embed.weight], "lr": token_lr, "base_lr": token_lr}) + if base_model.ve_shared.proj is not None: + scalar_params.append(base_model.ve_shared.proj.weight) + scalar_params.append(base_model.ve_shared.scale) + for s in base_model.ve_layer_scales: + scalar_params.append(s) + optimizer_tok = torch.optim.AdamW( + tok_params, + betas=(args.beta1, args.beta2), + eps=args.adam_eps, + weight_decay=args.embed_wd, + fused=True, + ) + optimizer_muon = Muon( + matrix_params, + lr=args.matrix_lr, + momentum=args.muon_momentum, + backend_steps=args.muon_backend_steps, + weight_decay=args.muon_wd, + ) + for group in optimizer_muon.param_groups: + group["base_lr"] = args.matrix_lr + optimizer_scalar = torch.optim.AdamW( + [{"params": scalar_params, "lr": args.scalar_lr, "base_lr": args.scalar_lr}], + betas=(args.beta1, args.beta2), + eps=args.adam_eps, + weight_decay=args.adam_wd, + fused=True, + ) + # Non-bank params that need manual all-reduce (replicated across GPUs) + replicated_params = list(optimizer_tok.param_groups[0]["params"]) + for pg in optimizer_tok.param_groups[1:]: + replicated_params.extend(pg["params"]) + replicated_params.extend(scalar_params) + + optimizer_head = None + if base_model.lm_head is not None: + optimizer_head = torch.optim.Adam( + [{"params": [base_model.lm_head.weight], "lr": args.head_lr, "base_lr": args.head_lr}], + betas=(args.beta1, args.beta2), + eps=args.adam_eps, + fused=True, + ) + replicated_params.append(base_model.lm_head.weight) + optimizers: list[torch.optim.Optimizer] = [optimizer_tok, optimizer_muon, optimizer_scalar] + if optimizer_head is not None: + optimizers.append(optimizer_head) + n_params = sum(p.numel() for p in base_model.parameters()) + mtp_params = sum(p.numel() for p in base_model.mtp_heads.parameters()) + log0(f"model_params:{n_params}") + log0(f"mtp_num_heads:{args.mtp_num_heads} mtp_loss_weight:{args.mtp_loss_weight} mtp_params:{mtp_params}") + xsa_layers = [i for i, b in enumerate(base_model.blocks) if b.attn.use_xsa] + log0(f"XSA:last_{args.xsa_last_n} active_layers:{xsa_layers}") + log0(f"world_size:{world_size} grad_accum_steps:{grad_accum_steps}") + log0("sdp_backends:cudnn=False flash=True mem_efficient=False math=False") + log0(f"attention_mode:gqa num_heads:{args.num_heads} num_kv_heads:{args.num_kv_heads}") + log0( + f"tie_embeddings:{args.tie_embeddings} embed_lr:{token_lr} " + f"head_lr:{args.head_lr if base_model.lm_head is not None else 0.0} " + f"matrix_lr:{args.matrix_lr} scalar_lr:{args.scalar_lr}" + ) + log0( + f"train_batch_tokens:{args.train_batch_tokens} train_seq_len:{args.train_seq_len} " + f"iterations:{args.iterations} warmup_steps:{args.warmup_steps} " + f"max_wallclock_seconds:{args.max_wallclock_seconds:.3f}" + ) + log0(f"seed:{args.seed}") + train_loader = DistributedTokenLoader(args.train_files, rank, world_size, device) + def zero_grad_all() -> None: + for opt in optimizers: + opt.zero_grad(set_to_none=True) + max_wallclock_ms = 1000.0 * args.max_wallclock_seconds if args.max_wallclock_seconds > 0 else None + def training_frac(step: int, elapsed_ms: float) -> float: + if max_wallclock_ms is None: + return step / max(args.iterations, 1) + return elapsed_ms / max(max_wallclock_ms, 1e-9) + def lr_mul(step: int, elapsed_ms: float) -> float: + # warmdown_frac > 0 takes priority; fallback to legacy warmdown_iters + if args.warmdown_frac > 0: + frac = training_frac(step, elapsed_ms) + if frac >= 1.0 - args.warmdown_frac: + return max((1.0 - frac) / args.warmdown_frac, 0.0) + return 1.0 + if args.warmdown_iters <= 0: + return 1.0 + if max_wallclock_ms is None: + warmdown_start = max(args.iterations - args.warmdown_iters, 0) + return max((args.iterations - step) / max(args.warmdown_iters, 1), 0.0) if warmdown_start <= step < args.iterations else 1.0 + step_ms = elapsed_ms / max(step, 1) + warmdown_ms = args.warmdown_iters * step_ms + remaining_ms = max(max_wallclock_ms - elapsed_ms, 0.0) + return remaining_ms / max(warmdown_ms, 1e-9) if remaining_ms <= warmdown_ms else 1.0 + if args.warmup_steps > 0: + initial_model_state = {name: tensor.detach().cpu().clone() for name, tensor in base_model.state_dict().items()} + initial_optimizer_states = [copy.deepcopy(opt.state_dict()) for opt in optimizers] + model.train() + for warmup_step in range(args.warmup_steps): + zero_grad_all() + for micro_step in range(grad_accum_steps): + x, y = train_loader.next_batch(args.train_batch_tokens, args.train_seq_len, grad_accum_steps) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + warmup_loss = model(x, y) + (warmup_loss * grad_scale).backward() + # All-reduce all grads for warmup (simple, not optimized) + if distributed: + for p in base_model.parameters(): + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG) + for opt in optimizers: + opt.step() + zero_grad_all() + if args.warmup_steps <= 20 or (warmup_step + 1) % 10 == 0 or warmup_step + 1 == args.warmup_steps: + log0(f"warmup_step:{warmup_step + 1}/{args.warmup_steps}") + base_model.load_state_dict(initial_model_state, strict=True) + for opt, state in zip(optimizers, initial_optimizer_states, strict=True): + opt.load_state_dict(state) + zero_grad_all() + train_loader = DistributedTokenLoader(args.train_files, rank, world_size, device) + swa_state: dict[str, Tensor] | None = None + swa_count = 0 + from collections import deque + lawa_queue: deque[dict[str, Tensor]] = deque(maxlen=args.lawa_k) + ema_state = {name: t.detach().float().clone() for name, t in base_model.state_dict().items()} + ema_decay = 0.997 + training_time_ms = 0.0 + stop_after_step: int | None = None + torch.cuda.synchronize() + t0 = time.perf_counter() + step = 0 + while True: + last_step = step == args.iterations or (stop_after_step is not None and step >= stop_after_step) + should_validate = last_step or (args.val_loss_every > 0 and step % args.val_loss_every == 0) + if should_validate: + torch.cuda.synchronize() + training_time_ms += 1000.0 * (time.perf_counter() - t0) + val_loss, val_bpb = eval_val( + args, + model, + rank, + world_size, + device, + grad_accum_steps, + val_tokens, + base_bytes_lut, + has_leading_space_lut, + is_boundary_token_lut, + ) + log0( + f"step:{step}/{args.iterations} val_loss:{val_loss:.4f} val_bpb:{val_bpb:.4f} " + f"train_time:{training_time_ms:.0f}ms step_avg:{training_time_ms / max(step, 1):.2f}ms" + ) + torch.cuda.synchronize() + t0 = time.perf_counter() + if last_step: + if stop_after_step is not None and step < args.iterations: + log0( + f"stopping_early: wallclock_cap train_time:{training_time_ms:.0f}ms " + f"step:{step}/{args.iterations}" + ) + break + elapsed_ms = training_time_ms + 1000.0 * (time.perf_counter() - t0) + scale = lr_mul(step, elapsed_ms) + if args.late_qat_threshold > 0 and scale < args.late_qat_threshold and not CastedLinear._qat_enabled: + CastedLinear._qat_enabled = True + log0(f"late_qat:enabled step:{step} scale:{scale:.4f}") + zero_grad_all() + train_loss = torch.zeros((), device=device) + for micro_step in range(grad_accum_steps): + x, y = train_loader.next_batch(args.train_batch_tokens, args.train_seq_len, grad_accum_steps) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + loss = model(x, y) + train_loss += loss.detach() + (loss * grad_scale).backward() + train_loss /= grad_accum_steps + frac = min(step / args.muon_momentum_warmup_steps, 1.0) if args.muon_momentum_warmup_steps > 0 else 1.0 + muon_momentum = (1 - frac) * args.muon_momentum_warmup_start + frac * args.muon_momentum + for group in optimizer_muon.param_groups: + group["momentum"] = muon_momentum + for opt in optimizers: + for group in opt.param_groups: + group["lr"] = group["base_lr"] * scale + if args.grad_clip_norm > 0: + torch.nn.utils.clip_grad_norm_(base_model.parameters(), args.grad_clip_norm) + # === 3-phase overlapped optimizer step === + # Phase 1: Launch async reduce-scatter for banks (biggest first) + optimizer_muon.launch_reduce_scatters() + # Phase 2: All-reduce non-bank grads + step Adam (while bank RS is in-flight) + if distributed: + for p in replicated_params: + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG) + optimizer_tok.step() + optimizer_scalar.step() + if optimizer_head is not None: + optimizer_head.step() + # Phase 3: Wait for RS, local NS5, all-gather (banks processed last) + optimizer_muon.step() + zero_grad_all() + # EMA update + with torch.no_grad(): + for name, t in base_model.state_dict().items(): + ema_state[name].mul_(ema_decay).add_(t.detach().float(), alpha=1.0 - ema_decay) + step += 1 + approx_training_time_ms = training_time_ms + 1000.0 * (time.perf_counter() - t0) + if args.swa_enabled and scale < 0.2 and step % args.swa_every == 0: + if swa_state is None: + swa_state = {name: t.detach().cpu().clone() for name, t in base_model.state_dict().items()} + swa_count = 1 + log0(f"swa:start step:{step}") + else: + for name, t in base_model.state_dict().items(): + swa_state[name] += t.detach().cpu() + swa_count += 1 + if args.lawa_enabled and step % args.lawa_freq == 0: + lawa_queue.append({name: t.detach().cpu().clone() for name, t in base_model.state_dict().items()}) + should_log_train = ( + args.train_log_every > 0 + and (step <= 10 or step % args.train_log_every == 0 or stop_after_step is not None) + ) + if should_log_train: + log0( + f"step:{step}/{args.iterations} train_loss:{train_loss.item():.4f} " + f"train_time:{approx_training_time_ms:.0f}ms step_avg:{approx_training_time_ms / step:.2f}ms" + ) + reached_cap = max_wallclock_ms is not None and approx_training_time_ms >= max_wallclock_ms + if distributed and max_wallclock_ms is not None: + reached_cap_tensor = torch.tensor(int(reached_cap), device=device) + dist.all_reduce(reached_cap_tensor, op=dist.ReduceOp.MAX) + reached_cap = bool(reached_cap_tensor.item()) + if stop_after_step is None and reached_cap: + stop_after_step = step + log0( + f"peak memory allocated: {torch.cuda.max_memory_allocated() // 1024 // 1024} MiB " + f"reserved: {torch.cuda.max_memory_reserved() // 1024 // 1024} MiB" + ) + # Apply weight averaging + if args.lawa_enabled and len(lawa_queue) > 1: + log0(f"lawa:applying LAWA averaging k={len(lawa_queue)}") + current_state = base_model.state_dict() + avg_state = {name: torch.zeros(t.shape, dtype=torch.float32, device='cpu') for name, t in current_state.items()} + for snap in lawa_queue: + for name in avg_state: + avg_state[name] += snap[name].float() + for name in avg_state: + avg_state[name] /= len(lawa_queue) + avg_state[name] = avg_state[name].to(dtype=current_state[name].dtype) + base_model.load_state_dict(avg_state, strict=True) + else: + log0("ema:applying EMA weights") + current_state = base_model.state_dict() + avg_state = {name: t.to(dtype=current_state[name].dtype) for name, t in ema_state.items()} + base_model.load_state_dict(avg_state, strict=True) + torch.cuda.synchronize() + t_diag = time.perf_counter() + diag_val_loss, diag_val_bpb = eval_val( + args, compiled_model, rank, world_size, device, grad_accum_steps, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + ) + torch.cuda.synchronize() + log0( + f"DIAGNOSTIC post_ema val_loss:{diag_val_loss:.4f} val_bpb:{diag_val_bpb:.4f} " + f"eval_time:{1000.0 * (time.perf_counter() - t_diag):.0f}ms" + ) + # AdamW TTT: fine-tune EMA model on val data BEFORE quantization + if args.ttt_enabled: + if distributed: + dist.barrier() + log0(f"ttt:start lr={args.ttt_lr} epochs={args.ttt_epochs} " + f"freeze_blocks={args.ttt_freeze_blocks} cosine_decay={args.ttt_cosine_decay}") + t_ttt = time.perf_counter() + ttt_adapt_adamw( + args, base_model, device, val_tokens, + rank=rank, world_size=world_size, log0=log0, + ) + torch.cuda.synchronize() + log0(f"ttt:elapsed={time.perf_counter() - t_ttt:.1f}s") + t_ttt_diag = time.perf_counter() + ttt_diag_loss, ttt_diag_bpb = eval_val( + args, base_model, rank, world_size, device, grad_accum_steps, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + ) + torch.cuda.synchronize() + log0(f"DIAGNOSTIC post_ttt val_loss:{ttt_diag_loss:.4f} val_bpb:{ttt_diag_bpb:.4f} " + f"eval_time:{1000.0 * (time.perf_counter() - t_ttt_diag):.0f}ms") + if distributed: + dist.barrier() + full_state_dict = base_model.state_dict() + export_sd = {k: v for k, v in full_state_dict.items() if "mtp_heads" not in k} + excluded_mtp = sum(int(t.numel()) for k, t in full_state_dict.items() if "mtp_heads" in k) + if excluded_mtp > 0: + log0(f"export_excluding_mtp_params:{excluded_mtp}") + if master_process: + torch.save(export_sd, "final_model.pt") + model_bytes = os.path.getsize("final_model.pt") + code_bytes = len(code.encode("utf-8")) + log0(f"Serialized model: {model_bytes} bytes") + log0(f"Code size: {code_bytes} bytes") + # Unbank 3D tensors into individual 2D tensors for quantization + sd_cpu = {k: v.detach().cpu() for k, v in export_sd.items()} + unbanked_sd = _unbank_state_dict(sd_cpu, args.num_layers) + # Full GPTQ: collect Hessians via a temporary non-banked model + log0(f"gptq:building non-banked model for Hessian collection...") + hessian_model = _HessianGPT( + vocab_size=args.vocab_size, num_layers=args.num_layers, model_dim=args.model_dim, + num_heads=args.num_heads, num_kv_heads=args.num_kv_heads, mlp_mult=args.mlp_mult, + tie_embeddings=args.tie_embeddings, logit_softcap=args.logit_softcap, + rope_base=args.rope_base, qk_gain_init=args.qk_gain_init, + bigram_vocab_size=args.bigram_vocab_size, bigram_dim=args.bigram_dim, + xsa_last_n=args.xsa_last_n, rope_dims=args.rope_dims, ln_scale=args.ln_scale, + ve_enabled=args.ve_enabled, ve_dim=args.ve_dim, ve_layers=args.ve_layers, + ).to(device).bfloat16() + for m in hessian_model.modules(): + if isinstance(m, CastedLinear): + m.float() + restore_low_dim_params_to_fp32(hessian_model) + # Load unbanked weights into the non-banked model + hessian_model.load_state_dict( + {k: v.to(device) for k, v in unbanked_sd.items() if k in hessian_model.state_dict()}, + strict=False, + ) + # Training-data calibration (proven -0.0007 BPP vs AR self-gen) + log0(f"gptq:collecting hessians from training data ({args.gptq_calib_batches} batches)...") + hessians = collect_hessians(hessian_model, train_loader, args, device, grad_accum_steps, num_batches=args.gptq_calib_batches) + log0(f"gptq:collected hessians for {len(hessians)} layers (training data)") + del hessian_model + torch.cuda.empty_cache() + quant_result, quant_meta = mixed_quantize_int6( + unbanked_sd, {"mlp", "attn"}, hessians=hessians, + matrix_bits=args.matrix_bits, embed_bits=args.embed_bits, + matrix_clip_sigmas=args.matrix_clip_sigmas, embed_clip_sigmas=args.embed_clip_sigmas, + ) + # NOVEL: Selective ±1 pruning by reconstruction error + # Sort ±1 quantized values by their reconstruction error (scale²), + # prune least-impactful first until artifact fits target size. + target_mb = float(os.environ.get("TARGET_MB", "15.9")) + code_bytes_est = len(code.encode("utf-8")) + ones_info = [] # (tensor_key, flat_idx, error) + for name, info in quant_meta.items(): + if not (isinstance(info, dict) and info.get("type") == "int6"): continue + qk, sk = name + ".q", name + ".scale" + if qk not in quant_result or sk not in quant_result: continue + q, s = quant_result[qk], quant_result[sk] + if s.ndim > 0: + ones_mask = (q.abs() == 1) + if ones_mask.any(): + row_idx = torch.arange(q.shape[0]).unsqueeze(1).expand_as(q)[ones_mask] + flat_idx = torch.arange(q.numel()).reshape(q.shape)[ones_mask] + errors = s.float()[row_idx].pow(2) + for fi, err in zip(flat_idx.tolist(), errors.tolist()): + ones_info.append((qk, fi, err)) + if ones_info: + ones_info.sort(key=lambda x: x[2]) + def _try_prune(n): + tmp = {k: v.clone() for k, v in quant_result.items()} + for i in range(min(n, len(ones_info))): + tmp[ones_info[i][0]].view(-1)[ones_info[i][1]] = 0 + buf = io.BytesIO(); torch.save({"w": tmp, "m": quant_meta}, buf) + return len(_compress_brotli(buf.getvalue())) + code_bytes_est, tmp + no_sz, _ = _try_prune(0) + target_bytes = int(target_mb * 1024 * 1024) + log0(f"selective_prune: {len(ones_info)} ±1 candidates, unpruned={no_sz/(1024*1024):.2f}MB target={target_mb}MB") + if no_sz <= target_bytes: + log0("selective_prune: already fits, no pruning needed") + else: + full_sz, _ = _try_prune(len(ones_info)) + log0(f"selective_prune: full ±1 prune={full_sz/(1024*1024):.2f}MB") + if full_sz > target_bytes: + log0("selective_prune: even full prune not enough, applying all") + _, quant_result = _try_prune(len(ones_info)) + else: + lo, hi = 0, len(ones_info) + while lo < hi: + mid = (lo + hi) // 2 + sz, _ = _try_prune(mid) + if sz <= target_bytes: hi = mid + else: lo = mid + 1 + log0(f"selective_prune: pruning {lo}/{len(ones_info)} ±1 values ({100*lo/len(ones_info):.1f}%) to fit {target_mb}MB") + _, quant_result = _try_prune(lo) + quant_buf = io.BytesIO() + torch.save({"w": quant_result, "m": quant_meta}, quant_buf) + quant_raw = quant_buf.getvalue() + quant_blob = _compress_brotli(quant_raw) + if master_process: + with open("final_model.int6.ptz", "wb") as f: + f.write(quant_blob) + quant_file_bytes = len(quant_blob) + code_bytes = len(code.encode("utf-8")) + log0(f"Serialized model int6+brotli+byteshuffle: {quant_file_bytes} bytes") + log0(f"Total submission size int6+brotli+byteshuffle: {quant_file_bytes + code_bytes} bytes") + if distributed: + dist.barrier() + with open("final_model.int6.ptz", "rb") as f: + quant_blob_disk = f.read() + quant_state = torch.load( + io.BytesIO(_decompress_brotli(quant_blob_disk)), + map_location="cpu", + weights_only=False, + ) + deq_unbanked = dequantize_mixed_int6(quant_state["w"], quant_state["m"], unbanked_sd) + # Re-bank the dequantized tensors + deq_state = _rebank_state_dict(deq_unbanked, args.num_layers, sd_cpu) + eval_model = GPT( + vocab_size=args.vocab_size, num_layers=args.num_layers, model_dim=args.model_dim, + num_heads=args.num_heads, num_kv_heads=args.num_kv_heads, mlp_mult=args.mlp_mult, + tie_embeddings=args.tie_embeddings, tied_embed_init_std=args.tied_embed_init_std, + logit_softcap=args.logit_softcap, rope_base=args.rope_base, qk_gain_init=args.qk_gain_init, + mtp_num_heads=0, mtp_loss_weight=0.0, + bigram_vocab_size=args.bigram_vocab_size, bigram_dim=args.bigram_dim, + xsa_last_n=args.xsa_last_n, + rope_dims=args.rope_dims, ln_scale=args.ln_scale, dtg=args.dtg_enabled, + ve_enabled=args.ve_enabled, ve_dim=args.ve_dim, ve_layers=args.ve_layers, + gated_attention=args.gated_attention, value_residual=args.value_residual, + skip_gates_enabled=args.skip_gates_enabled, + ).to(device).bfloat16() + eval_model.qo_bank.data = eval_model.qo_bank.data.float() + eval_model.kv_bank.data = eval_model.kv_bank.data.float() + eval_model.mlp_up_bank.data = eval_model.mlp_up_bank.data.float() + eval_model.mlp_down_bank.data = eval_model.mlp_down_bank.data.float() + for m in eval_model.modules(): + if isinstance(m, CastedLinear): + m.float() + restore_low_dim_params_to_fp32(eval_model) + eval_model.load_state_dict(deq_state, strict=True) + compiled_eval = torch.compile(eval_model, dynamic=False, fullgraph=True) + torch.cuda.synchronize() + t_qeval = time.perf_counter() + q_val_loss, q_val_bpb = eval_val( + args, compiled_eval, rank, world_size, device, grad_accum_steps, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + eval_seq_len=effective_eval_seq_len, + ) + torch.cuda.synchronize() + log0( + f"final_int6_roundtrip val_loss:{q_val_loss:.4f} val_bpb:{q_val_bpb:.4f} " + f"eval_time:{1000.0 * (time.perf_counter() - t_qeval):.0f}ms" + ) + log0(f"final_int6_roundtrip_exact val_loss:{q_val_loss:.8f} val_bpb:{q_val_bpb:.8f}") + sw_seq_len = effective_eval_seq_len + if args.eval_stride > 0 and args.eval_stride < sw_seq_len: + torch.cuda.synchronize() + t_slide = time.perf_counter() + sw_val_loss, sw_val_bpb = eval_val_sliding( + args, eval_model, rank, world_size, device, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + stride=args.eval_stride, + eval_seq_len=sw_seq_len, + ) + torch.cuda.synchronize() + log0( + f"final_int6_sliding_window val_loss:{sw_val_loss:.4f} val_bpb:{sw_val_bpb:.4f} " + f"stride:{args.eval_stride} eval_time:{1000.0 * (time.perf_counter() - t_slide):.0f}ms" + ) + log0(f"final_int6_sliding_window_exact val_loss:{sw_val_loss:.8f} val_bpb:{sw_val_bpb:.8f}") + log0(f"final_int8_zlib_roundtrip_exact val_loss:{sw_val_loss:.8f} val_bpb:{sw_val_bpb:.8f}") + if args.eval_stride != 64 and 64 < sw_seq_len: + torch.cuda.synchronize() + t_slide64 = time.perf_counter() + sw64_val_loss, sw64_val_bpb = eval_val_sliding( + args, eval_model, rank, world_size, device, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + stride=64, + eval_seq_len=sw_seq_len, + ) + torch.cuda.synchronize() + log0( + f"final_int6_sliding_window_s64 val_loss:{sw64_val_loss:.4f} val_bpb:{sw64_val_bpb:.4f} " + f"stride:64 eval_time:{1000.0 * (time.perf_counter() - t_slide64):.0f}ms" + ) + log0(f"final_int6_sliding_window_s64_exact val_loss:{sw64_val_loss:.8f} val_bpb:{sw64_val_bpb:.8f}") + log0(f"final_int8_zlib_roundtrip_exact val_loss:{sw64_val_loss:.8f} val_bpb:{sw64_val_bpb:.8f}") + # --- Causal SLOT: per-batch delta optimization using ONLY context (already-scored) positions --- + if args.slot_enabled and args.eval_stride > 0 and args.eval_stride < sw_seq_len: + try: + slot_stride = args.eval_stride + seq_s = sw_seq_len + total_tok = val_tokens.numel() - 1 + ws_list = [ws for ws in range(0, total_tok, slot_stride) if min(ws + seq_s, total_tok) - ws >= 1] + my_s = (len(ws_list) * rank) // world_size + my_e = (len(ws_list) * (rank + 1)) // world_size + my_ws = ws_list[my_s:my_e] + num_batches = (len(my_ws) + 31) // 32 + sl_loss = torch.zeros((), device=device, dtype=torch.float64) + sl_tc = torch.zeros((), device=device, dtype=torch.float64) + sl_bc = torch.zeros((), device=device, dtype=torch.float64) + torch.cuda.synchronize() + t_slot = time.perf_counter() + eval_model.eval() + log0(f"causal_slot:start lr={args.slot_lr} steps={args.slot_steps} stride={slot_stride} " + f"windows={len(my_ws)} batches={num_batches}") + for batch_idx, bi in enumerate(range(0, len(my_ws), 32)): + bws = my_ws[bi:bi+32]; bsz = len(bws) + xb = torch.zeros(bsz, seq_s, dtype=torch.int64, device=device) + yb = torch.zeros(bsz, seq_s, dtype=torch.int64, device=device) + wls = [] + for i, ws in enumerate(bws): + end = min(ws + seq_s, total_tok); wl = end - ws; wls.append(wl) + ct = val_tokens[ws:end+1].to(dtype=torch.int64, device=device) + xb[i,:wl] = ct[:-1]; yb[i,:wl] = ct[1:] + # Frozen forward: get hidden states + with torch.no_grad(), torch.autocast(device_type="cuda", dtype=torch.bfloat16): + H = eval_model.forward_hidden(xb) + H = H.detach().float() + # Build context mask (already-scored positions) and score mask (new positions) + context_mask = torch.zeros(bsz, seq_s, dtype=torch.bool, device=device) + has_context = False + for i, ws in enumerate(bws): + wl = wls[i]; s = 0 if ws == 0 else max(wl - slot_stride, 0) + if s > 0: + context_mask[i, :s] = True + has_context = True + # Optimize delta on context positions only (CAUSAL: no future info) + delta = torch.zeros(1, 1, H.shape[-1], device=device, dtype=H.dtype, requires_grad=True) + if has_context: + sopt = torch.optim.AdamW([delta], lr=args.slot_lr, betas=(0.3, 0.9), weight_decay=1e-8, eps=1e-5) + for _ in range(args.slot_steps): + sopt.zero_grad() + lg = eval_model.compute_logits((H + delta).to(torch.bfloat16)).float() + nll_all = F.cross_entropy(lg.reshape(-1, lg.size(-1)), yb.reshape(-1), reduction="none").reshape(bsz, seq_s) + ctx_nll = nll_all[context_mask] + if ctx_nll.numel() > 0: + loss_c = ctx_nll.mean() + loss_c.backward() + sopt.step() + # Score new positions with adapted delta + with torch.no_grad(): + lg = eval_model.compute_logits((H + delta.detach()).to(torch.bfloat16)).float() + nll = F.cross_entropy(lg.reshape(-1, lg.size(-1)), yb.reshape(-1), reduction="none").reshape(bsz, seq_s) + for i, ws in enumerate(bws): + wl = wls[i]; s = 0 if ws == 0 else max(wl - slot_stride, 0) + sl_loss += nll[i, s:wl].to(torch.float64).sum(); sl_tc += float(wl - s) + tgt, prev = yb[i, s:wl], xb[i, s:wl] + tb = base_bytes_lut[tgt].to(torch.float64) + tb += (has_leading_space_lut[tgt] & ~is_boundary_token_lut[prev]).to(torch.float64) + sl_bc += tb.sum() + if batch_idx % 500 == 0 or batch_idx == num_batches - 1: + log0(f" causal_slot:batch {batch_idx+1}/{num_batches} " + f"time:{time.perf_counter()-t_slot:.1f}s"); sys.stdout.flush() + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(sl_loss, op=dist.ReduceOp.SUM) + dist.all_reduce(sl_tc, op=dist.ReduceOp.SUM) + dist.all_reduce(sl_bc, op=dist.ReduceOp.SUM) + sv_loss = (sl_loss / sl_tc).item() + sv_bpb = sv_loss / math.log(2.0) * (sl_tc.item() / sl_bc.item()) + torch.cuda.synchronize() + log0(f"final_causal_slot val_loss:{sv_loss:.4f} val_bpb:{sv_bpb:.4f} time:{1000*(time.perf_counter()-t_slot):.0f}ms") + log0(f"final_causal_slot_exact val_loss:{sv_loss:.8f} val_bpb:{sv_bpb:.8f}") + log0(f"final_int8_zlib_roundtrip_exact val_loss:{sv_loss:.8f} val_bpb:{sv_bpb:.8f}") + except Exception as e: + import traceback; log0(f"causal_slot:ERROR {e}"); traceback.print_exc(); sys.stdout.flush() + if distributed: + dist.destroy_process_group() + sys.exit(0) +if __name__ == "__main__": + main() + +==================================================================================================== +Running Python 3.12.3 (main, Nov 6 2025, 13:44:16) [GCC 13.3.0] +Running PyTorch 2.9.1+cu128 +Mon Apr 6 05:02:53 2026 ++-----------------------------------------------------------------------------------------+ +| NVIDIA-SMI 580.126.09 Driver Version: 580.126.09 CUDA Version: 13.0 | ++-----------------------------------------+------------------------+----------------------+ +| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC | +| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. | +| | | MIG M. | +|=========================================+========================+======================| +| 0 NVIDIA H100 80GB HBM3 On | 00000000:19:00.0 Off | 0 | +| N/A 33C P0 117W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 1 NVIDIA H100 80GB HBM3 On | 00000000:3B:00.0 Off | 0 | +| N/A 30C P0 117W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 2 NVIDIA H100 80GB HBM3 On | 00000000:4C:00.0 Off | 0 | +| N/A 29C P0 114W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 3 NVIDIA H100 80GB HBM3 On | 00000000:5D:00.0 Off | 0 | +| N/A 32C P0 117W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 4 NVIDIA H100 80GB HBM3 On | 00000000:9B:00.0 Off | 0 | +| N/A 33C P0 116W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 5 NVIDIA H100 80GB HBM3 On | 00000000:BB:00.0 Off | 0 | +| N/A 32C P0 117W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 6 NVIDIA H100 80GB HBM3 On | 00000000:CB:00.0 Off | 0 | +| N/A 32C P0 116W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 7 NVIDIA H100 80GB HBM3 On | 00000000:DB:00.0 Off | 0 | +| N/A 29C P0 110W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ + ++-----------------------------------------------------------------------------------------+ +| Processes: | +| GPU GI CI PID Type Process name GPU Memory | +| ID ID Usage | +|=========================================================================================| +| No running processes found | ++-----------------------------------------------------------------------------------------+ + +==================================================================================================== +val_bpb:enabled tokenizer_kind=sentencepiece tokenizer_path=./data/tokenizers/fineweb_8192_bpe.model +train_loader:dataset:fineweb10B_sp8192 train_shards:128 +val_loader:shards pattern=./data/datasets/fineweb10B_sp8192/fineweb_val_*.bin tokens:40540160 +model_params:35941976 +mtp_num_heads:0 mtp_loss_weight:0.2 mtp_params:0 +XSA:last_11 active_layers:[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10] +world_size:8 grad_accum_steps:1 +sdp_backends:cudnn=False flash=True mem_efficient=False math=False +attention_mode:gqa num_heads:8 num_kv_heads:4 +tie_embeddings:True embed_lr:0.035 head_lr:0.0 matrix_lr:0.025 scalar_lr:0.025 +train_batch_tokens:786432 train_seq_len:2048 iterations:20000 warmup_steps:20 max_wallclock_seconds:600.000 +seed:1337 +warmup_step:1/20 +warmup_step:2/20 +warmup_step:3/20 +warmup_step:4/20 +warmup_step:5/20 +warmup_step:6/20 +warmup_step:7/20 +warmup_step:8/20 +warmup_step:9/20 +warmup_step:10/20 +warmup_step:11/20 +warmup_step:12/20 +warmup_step:13/20 +warmup_step:14/20 +warmup_step:15/20 +warmup_step:16/20 +warmup_step:17/20 +warmup_step:18/20 +warmup_step:19/20 +warmup_step:20/20 +step:0/20000 val_loss:9.0006 val_bpb:3.4844 train_time:0ms step_avg:0.02ms +step:1/20000 train_loss:9.0023 train_time:146ms step_avg:146.20ms +step:2/20000 train_loss:9.7748 train_time:176ms step_avg:87.96ms +step:3/20000 train_loss:10.0687 train_time:272ms step_avg:90.69ms +step:4/20000 train_loss:9.5448 train_time:369ms step_avg:92.25ms +step:5/20000 train_loss:8.8965 train_time:466ms step_avg:93.14ms +step:6/20000 train_loss:8.6403 train_time:563ms step_avg:93.85ms +step:7/20000 train_loss:8.2936 train_time:662ms step_avg:94.55ms +step:8/20000 train_loss:7.8568 train_time:761ms step_avg:95.11ms +step:9/20000 train_loss:7.8099 train_time:861ms step_avg:95.63ms +step:10/20000 train_loss:7.6015 train_time:961ms step_avg:96.10ms +step:100/20000 train_loss:4.4639 train_time:9938ms step_avg:99.38ms +step:200/20000 train_loss:3.7779 train_time:19903ms step_avg:99.52ms +step:300/20000 train_loss:3.7487 train_time:29877ms step_avg:99.59ms +step:400/20000 train_loss:3.4990 train_time:39850ms step_avg:99.63ms +step:500/20000 train_loss:3.3475 train_time:49805ms step_avg:99.61ms +step:600/20000 train_loss:3.4423 train_time:59826ms step_avg:99.71ms +step:700/20000 train_loss:3.2661 train_time:69865ms step_avg:99.81ms +step:800/20000 train_loss:3.3674 train_time:79904ms step_avg:99.88ms +step:900/20000 train_loss:3.2469 train_time:89958ms step_avg:99.95ms +step:1000/20000 train_loss:3.0726 train_time:99961ms step_avg:99.96ms +step:1000/20000 val_loss:3.2717 val_bpb:1.2666 train_time:100027ms step_avg:100.03ms +step:1100/20000 train_loss:3.2406 train_time:110000ms step_avg:100.00ms +step:1200/20000 train_loss:3.3439 train_time:120024ms step_avg:100.02ms +step:1300/20000 train_loss:3.3333 train_time:130059ms step_avg:100.05ms +step:1400/20000 train_loss:3.2681 train_time:140103ms step_avg:100.07ms +step:1500/20000 train_loss:3.2788 train_time:150064ms step_avg:100.04ms +step:1600/20000 train_loss:3.3263 train_time:160103ms step_avg:100.06ms +step:1700/20000 train_loss:3.1368 train_time:170140ms step_avg:100.08ms +step:1800/20000 train_loss:3.1615 train_time:180159ms step_avg:100.09ms +step:1900/20000 train_loss:3.1077 train_time:190131ms step_avg:100.07ms +step:2000/20000 train_loss:3.1182 train_time:200150ms step_avg:100.08ms +step:2000/20000 val_loss:3.1450 val_bpb:1.2175 train_time:200216ms step_avg:100.11ms +step:2100/20000 train_loss:3.2023 train_time:210173ms step_avg:100.08ms +step:2200/20000 train_loss:3.1164 train_time:220183ms step_avg:100.08ms +step:2300/20000 train_loss:3.1767 train_time:230227ms step_avg:100.10ms +step:2400/20000 train_loss:3.1473 train_time:240202ms step_avg:100.08ms +step:2500/20000 train_loss:3.2572 train_time:250217ms step_avg:100.09ms +step:2600/20000 train_loss:3.0968 train_time:260228ms step_avg:100.09ms +step:2700/20000 train_loss:3.1873 train_time:270241ms step_avg:100.09ms +step:2800/20000 train_loss:3.1571 train_time:280262ms step_avg:100.09ms +step:2900/20000 train_loss:3.0434 train_time:290221ms step_avg:100.08ms +step:3000/20000 train_loss:2.9684 train_time:300222ms step_avg:100.07ms +step:3000/20000 val_loss:3.0767 val_bpb:1.1911 train_time:300288ms step_avg:100.10ms +step:3100/20000 train_loss:3.1276 train_time:310240ms step_avg:100.08ms +step:3200/20000 train_loss:3.1267 train_time:320248ms step_avg:100.08ms +step:3300/20000 train_loss:3.2289 train_time:330272ms step_avg:100.08ms +step:3400/20000 train_loss:3.1007 train_time:340296ms step_avg:100.09ms +step:3500/20000 train_loss:3.1125 train_time:350308ms step_avg:100.09ms +step:3600/20000 train_loss:3.0020 train_time:360313ms step_avg:100.09ms +step:3700/20000 train_loss:3.0256 train_time:370329ms step_avg:100.09ms +step:3800/20000 train_loss:3.0694 train_time:380280ms step_avg:100.07ms +step:3900/20000 train_loss:3.1647 train_time:390285ms step_avg:100.07ms +step:4000/20000 train_loss:3.0289 train_time:400276ms step_avg:100.07ms +step:4000/20000 val_loss:3.0143 val_bpb:1.1669 train_time:400342ms step_avg:100.09ms +step:4100/20000 train_loss:3.0272 train_time:410277ms step_avg:100.07ms +step:4200/20000 train_loss:2.9891 train_time:420267ms step_avg:100.06ms +step:4300/20000 train_loss:3.0931 train_time:430208ms step_avg:100.05ms +step:4400/20000 train_loss:3.0590 train_time:440210ms step_avg:100.05ms +step:4500/20000 train_loss:3.0509 train_time:450205ms step_avg:100.05ms +step:4600/20000 train_loss:2.9966 train_time:460216ms step_avg:100.05ms +step:4700/20000 train_loss:3.0050 train_time:470143ms step_avg:100.03ms +step:4800/20000 train_loss:2.9482 train_time:480144ms step_avg:100.03ms +step:4900/20000 train_loss:2.8632 train_time:490122ms step_avg:100.02ms +step:5000/20000 train_loss:2.8829 train_time:500120ms step_avg:100.02ms +step:5000/20000 val_loss:2.9418 val_bpb:1.1389 train_time:500184ms step_avg:100.04ms +step:5100/20000 train_loss:3.0965 train_time:510110ms step_avg:100.02ms +swa:start step:5200 +step:5200/20000 train_loss:2.9110 train_time:520044ms step_avg:100.01ms +step:5300/20000 train_loss:2.9665 train_time:530259ms step_avg:100.05ms +late_qat:enabled step:5395 scale:0.1500 +step:5400/20000 train_loss:2.9925 train_time:540415ms step_avg:100.08ms +step:5500/20000 train_loss:2.8732 train_time:550568ms step_avg:100.10ms +step:5600/20000 train_loss:2.9359 train_time:560712ms step_avg:100.13ms +step:5700/20000 train_loss:2.8742 train_time:570816ms step_avg:100.14ms +step:5800/20000 train_loss:2.6129 train_time:580966ms step_avg:100.17ms +step:5900/20000 train_loss:2.7603 train_time:591133ms step_avg:100.19ms +step:5988/20000 val_loss:2.8500 val_bpb:1.1033 train_time:600161ms step_avg:100.23ms +stopping_early: wallclock_cap train_time:600161ms step:5988/20000 +peak memory allocated: 26628 MiB reserved: 26750 MiB +ema:applying EMA weights +DIAGNOSTIC post_ema val_loss:2.8463 val_bpb:1.1019 eval_time:1596ms +ttt:start lr=0.0005 epochs=6 freeze_blocks=2 cosine_decay=True +ttt_adamw:params trainable=35937864 frozen=4112 +ttt_adamw:epoch 1/6 loss:2.9067 time:14.2s +ttt_adamw:epoch 2/6 loss:2.8341 time:28.2s +ttt_adamw:epoch 3/6 loss:2.8110 time:42.2s +ttt_adamw:epoch 4/6 loss:2.7922 time:56.2s +ttt_adamw:epoch 5/6 loss:2.7761 time:70.2s +ttt_adamw:epoch 6/6 loss:2.7649 time:84.2s +ttt_adamw:done elapsed=84.2s +ttt:elapsed=84.2s +DIAGNOSTIC post_ttt val_loss:2.7586 val_bpb:1.0680 eval_time:4730ms +Serialized model: 135395695 bytes +Code size: 116724 bytes +gptq:building non-banked model for Hessian collection... +gptq:collecting hessians from training data (256 batches)... +gptq:collected hessians for 67 layers (training data) +selective_prune: 9816017 ±1 candidates, unpruned=14.42MB target=15.9MB +selective_prune: already fits, no pruning needed +Serialized model int6+brotli+byteshuffle: 15000558 bytes +Total submission size int6+brotli+byteshuffle: 15117282 bytes +final_int6_roundtrip val_loss:2.8251 val_bpb:1.0937 eval_time:18924ms +final_int6_roundtrip_exact val_loss:2.82513396 val_bpb:1.09369739 +final_int6_sliding_window val_loss:2.7877 val_bpb:1.0792 stride:64 eval_time:92121ms +final_int6_sliding_window_exact val_loss:2.78767945 val_bpb:1.07920213 +final_int8_zlib_roundtrip_exact val_loss:2.78767945 val_bpb:1.07920213 diff --git a/records/track_10min_16mb/2026-04-06_SP8192_PreQuantTTT_SDClip_SkipGates/train_seed2025.log b/records/track_10min_16mb/2026-04-06_SP8192_PreQuantTTT_SDClip_SkipGates/train_seed2025.log new file mode 100644 index 0000000000..8b091d8d11 --- /dev/null +++ b/records/track_10min_16mb/2026-04-06_SP8192_PreQuantTTT_SDClip_SkipGates/train_seed2025.log @@ -0,0 +1,2559 @@ +from __future__ import annotations +import copy +import glob +import io +import lzma +import math +import os +import random +import subprocess +import sys +import time +import uuid +import zlib +from pathlib import Path +try: + import zstandard + _COMPRESSOR = "zstd" +except ImportError: + _COMPRESSOR = "zlib" +import numpy as np +import sentencepiece as spm +import torch +import torch.distributed as dist +import torch.nn.functional as F +from torch import Tensor, nn +from torch.nn.parallel import DistributedDataParallel as DDP +try: + from flash_attn_interface import flash_attn_func as flash_attn_3_func + HAS_FA3 = True +except ImportError: + HAS_FA3 = False +class Hyperparameters: + data_path = os.environ.get("DATA_PATH", "./data/datasets/fineweb10B_sp8192") + train_files = os.path.join(data_path, "fineweb_train_*.bin") + val_files = os.path.join(data_path, "fineweb_val_*.bin") + tokenizer_path = os.environ.get("TOKENIZER_PATH", "./data/tokenizers/fineweb_8192_bpe.model") + run_id = os.environ.get("RUN_ID", str(uuid.uuid4())) + seed = int(os.environ.get("SEED", 1337)) + val_batch_size = int(os.environ.get("VAL_BATCH_SIZE", 524_288)) + val_loss_every = int(os.environ.get("VAL_LOSS_EVERY", 4000)) + train_log_every = int(os.environ.get("TRAIN_LOG_EVERY", 500)) + iterations = int(os.environ.get("ITERATIONS", 20000)) + warmdown_iters = int(os.environ.get("WARMDOWN_ITERS", 3500)) # legacy; overridden by warmdown_frac if > 0 + warmdown_frac = float(os.environ.get("WARMDOWN_FRAC", 0.667)) # fraction of total time for warmdown + warmup_steps = int(os.environ.get("WARMUP_STEPS", 20)) + train_batch_tokens = int(os.environ.get("TRAIN_BATCH_TOKENS", 786_432)) + train_seq_len = int(os.environ.get("TRAIN_SEQ_LEN", 2048)) + eval_seq_len = int(os.environ.get("EVAL_SEQ_LEN", 2048)) + max_wallclock_seconds = float(os.environ.get("MAX_WALLCLOCK_SECONDS", 600.0)) + qk_gain_init = float(os.environ.get("QK_GAIN_INIT", 4.0)) + vocab_size = int(os.environ.get("VOCAB_SIZE", 8192)) + num_layers = int(os.environ.get("NUM_LAYERS", 11)) + num_kv_heads = int(os.environ.get("NUM_KV_HEADS", 4)) + model_dim = int(os.environ.get("MODEL_DIM", 512)) + num_heads = int(os.environ.get("NUM_HEADS", 8)) + mlp_mult = float(os.environ.get("MLP_MULT", 4.0)) + tie_embeddings = bool(int(os.environ.get("TIE_EMBEDDINGS", "1"))) + rope_base = float(os.environ.get("ROPE_BASE", 10000.0)) + logit_softcap = float(os.environ.get("LOGIT_SOFTCAP", 30.0)) + embed_lr = float(os.environ.get("EMBED_LR", 0.6)) + head_lr = float(os.environ.get("HEAD_LR", 0.008)) + tied_embed_lr = float(os.environ.get("TIED_EMBED_LR", 0.035)) + tied_embed_init_std = float(os.environ.get("TIED_EMBED_INIT_STD", 0.005)) + matrix_lr = float(os.environ.get("MATRIX_LR", 0.025)) + scalar_lr = float(os.environ.get("SCALAR_LR", 0.025)) + muon_momentum = float(os.environ.get("MUON_MOMENTUM", 0.99)) + muon_backend_steps = int(os.environ.get("MUON_BACKEND_STEPS", 5)) + muon_momentum_warmup_start = float(os.environ.get("MUON_MOMENTUM_WARMUP_START", 0.92)) + muon_momentum_warmup_steps = int(os.environ.get("MUON_MOMENTUM_WARMUP_STEPS", 1500)) + beta1 = float(os.environ.get("BETA1", 0.9)) + beta2 = float(os.environ.get("BETA2", 0.95)) + adam_eps = float(os.environ.get("ADAM_EPS", 1e-8)) + grad_clip_norm = float(os.environ.get("GRAD_CLIP_NORM", 0.3)) + eval_stride = int(os.environ.get("EVAL_STRIDE", 64)) + mtp_num_heads = int(os.environ.get("MTP_NUM_HEADS", 0)) + mtp_loss_weight = float(os.environ.get("MTP_LOSS_WEIGHT", 0.2)) + muon_beta2 = float(os.environ.get("MUON_BETA2", 0.95)) + swa_enabled = bool(int(os.environ.get("SWA_ENABLED", "1"))) + swa_every = int(os.environ.get("SWA_EVERY", 50)) + lawa_enabled = bool(int(os.environ.get("LAWA_ENABLED", "0"))) + lawa_k = int(os.environ.get("LAWA_K", 10)) + lawa_freq = int(os.environ.get("LAWA_FREQ", 100)) + muon_wd = float(os.environ.get("MUON_WD", 0.085)) + adam_wd = float(os.environ.get("ADAM_WD", 0.02)) + embed_wd = float(os.environ.get("EMBED_WD", 0.085)) + qat_enabled = bool(int(os.environ.get("QAT_ENABLED", "0"))) + bigram_vocab_size = int(os.environ.get("BIGRAM_VOCAB_SIZE", 0)) # disabled: SP8192 replaces BigramHash + bigram_dim = int(os.environ.get("BIGRAM_DIM", 128)) + trigram_enabled = bool(int(os.environ.get("TRIGRAM", "0"))) # TrigramHash (off by default, risky) + xsa_last_n = int(os.environ.get("XSA_LAST_N", 11)) # XSA on ALL layers (our novel contribution) + rope_dims = int(os.environ.get("ROPE_DIMS", 16)) + ln_scale = bool(int(os.environ.get("LN_SCALE", "1"))) + dtg_enabled = bool(int(os.environ.get("DTG_ENABLED", "0"))) + late_qat_threshold = float(os.environ.get("LATE_QAT_THRESHOLD", 0.15)) + ve_enabled = bool(int(os.environ.get("VE_ENABLED", "0"))) + ve_dim = int(os.environ.get("VE_DIM", 128)) + ve_layers = os.environ.get("VE_LAYERS", "9,10") + gated_attention = bool(int(os.environ.get("GATED_ATTENTION", "0"))) + value_residual = bool(int(os.environ.get("VALUE_RESIDUAL", "0"))) # VRL with sigmoid gates (off by default, risky) + skip_gates_enabled = bool(int(os.environ.get("SKIP_GATES_ENABLED", "1"))) # learned sigmoid gates on U-Net skips + # GPTQ calibration + gptq_calib_batches = int(os.environ.get("GPTQ_CALIB_BATCHES", 256)) + gptq_block_size = int(os.environ.get("GPTQ_BLOCK_SIZE", 128)) + # SDClip: clip threshold = k * std(row) instead of percentile search + matrix_bits = int(os.environ.get("MATRIX_BITS", 6)) + embed_bits = int(os.environ.get("EMBED_BITS", 8)) + matrix_clip_sigmas = float(os.environ.get("MATRIX_CLIP_SIGMAS", 12.85)) # k for int6 matrices + embed_clip_sigmas = float(os.environ.get("EMBED_CLIP_SIGMAS", 20.0)) # k for int8 embeddings + # TTT: AdamW test-time training (pre-quantization, on EMA model) + ttt_enabled = bool(int(os.environ.get("TTT_ENABLED", "1"))) + ttt_lr = float(os.environ.get("TTT_LR", 0.0005)) + ttt_epochs = int(os.environ.get("TTT_EPOCHS", 6)) + ttt_batch_seqs = int(os.environ.get("TTT_BATCH_SEQS", 32)) + ttt_freeze_blocks = int(os.environ.get("TTT_FREEZE_BLOCKS", 2)) + ttt_grad_clip = float(os.environ.get("TTT_GRAD_CLIP", 1.0)) + ttt_cosine_decay = bool(int(os.environ.get("TTT_COSINE_DECAY", "1"))) + # Causal SLOT: per-batch delta optimization at last hidden layer (post-quantization) + slot_enabled = bool(int(os.environ.get("SLOT_ENABLED", "0"))) + slot_lr = float(os.environ.get("SLOT_LR", 0.005)) + slot_steps = int(os.environ.get("SLOT_STEPS", 8)) + +# --- Batched Newton-Schulz orthogonalization --- + +def zeropower_via_newtonschulz5(G: Tensor, steps: int = 5, eps: float = 1e-7) -> Tensor: + """Batched Newton-Schulz orthogonalization. G: (B,M,N) or (M,N).""" + a, b, c = (3.4445, -4.7750, 2.0315) + was_2d = G.ndim == 2 + if was_2d: + G = G.unsqueeze(0) + X = G.bfloat16() + transposed = X.size(-2) > X.size(-1) + if transposed: + X = X.mT + X = X / (X.norm(dim=(-2, -1), keepdim=True) + eps) + for _ in range(steps): + A = X @ X.mT + B = b * A + c * (A @ A) + X = a * X + B @ X + if transposed: + X = X.mT + if was_2d: + X = X.squeeze(0) + return X + +# --- Parallel Muon optimizer --- + +class Muon(torch.optim.Optimizer): + """Parallel Muon: post-backward reduce-scatter -> local NS5 -> all-gather. + + No DDP for bank params. After backward, this optimizer: + 1. Launches async reduce-scatter for all banks (biggest first) + 2. Returns control so Adam can step on small params while RS is in-flight + 3. Waits for each RS, runs local NS5 on the shard, launches async all-gather + 4. Each all-gather overlaps with next bank's NS5 + """ + def __init__(self, params, lr: float, momentum: float, backend_steps: int, + nesterov: bool = True, weight_decay: float = 0.0): + super().__init__( + params, + dict(lr=lr, momentum=momentum, backend_steps=backend_steps, + nesterov=nesterov, weight_decay=weight_decay), + ) + self._built = False + + def _build(self): + self._distributed = dist.is_available() and dist.is_initialized() + self._world_size = dist.get_world_size() if self._distributed else 1 + self._rank = dist.get_rank() if self._distributed else 0 + ws = self._world_size + + self._bank_meta = [] + for group in self.param_groups: + for p in group["params"]: + B = p.shape[0] + padded_B = ((B + ws - 1) // ws) * ws + shard_B = padded_B // ws + tail = p.shape[1:] + dev = p.device + self._bank_meta.append({ + 'p': p, + 'B': B, + 'padded_grad': torch.zeros(padded_B, *tail, device=dev, dtype=torch.bfloat16), + 'shard': torch.zeros(shard_B, *tail, device=dev, dtype=torch.bfloat16), + 'shard_mom': torch.zeros(shard_B, *tail, device=dev, dtype=torch.bfloat16), + 'full_update': torch.zeros(padded_B, *tail, device=dev, dtype=torch.bfloat16), + 'scale': max(1, p.shape[-2] / p.shape[-1]) ** 0.5, + }) + # Sort by size descending -- launch biggest reduce-scatters first + self._bank_meta.sort(key=lambda m: -m['p'].numel()) + self._built = True + + def launch_reduce_scatters(self): + """Phase 1: launch async reduce-scatter for all banks. Call right after backward.""" + if not self._built: + self._build() + if not self._distributed: + return + self._rs_futures = [] + for m in self._bank_meta: + p = m['p'] + if p.grad is None: + self._rs_futures.append(None) + continue + pg = m['padded_grad'] + pg[:m['B']].copy_(p.grad.bfloat16()) + if pg.shape[0] > m['B']: + pg[m['B']:].zero_() + fut = dist.reduce_scatter_tensor(m['shard'], pg, op=dist.ReduceOp.AVG, async_op=True) + self._rs_futures.append(fut) + + @torch.no_grad() + def step(self, closure=None): + """Phase 3: wait for RS, local NS5, all-gather. Call AFTER Adam steps.""" + loss = None + if closure is not None: + with torch.enable_grad(): + loss = closure() + + if not self._built: + self._build() + + for group in self.param_groups: + lr = group["lr"] + momentum = group["momentum"] + backend_steps = group["backend_steps"] + nesterov = group["nesterov"] + wd = group.get("weight_decay", 0.0) + + prev_ag_handle = None + prev_m = None + + sharded = self._distributed and hasattr(self, '_rs_futures') + + for i, m in enumerate(self._bank_meta): + p = m['p'] + if p.grad is None: + continue + + if prev_ag_handle is not None: + prev_ag_handle.wait() + pp = prev_m['p'] + upd = prev_m['full_update'][:prev_m['B']] + if wd > 0.0: + pp.data.mul_(1.0 - lr * wd) + pp.add_(upd.to(dtype=pp.dtype), alpha=-lr * prev_m['scale']) + + if sharded and self._rs_futures[i] is not None: + self._rs_futures[i].wait() + g = m['shard'] + buf = m['shard_mom'] + else: + g = p.grad.bfloat16() + state = self.state[p] + if "momentum_buffer" not in state: + state["momentum_buffer"] = torch.zeros_like(g) + buf = state["momentum_buffer"] + + buf.mul_(momentum).add_(g) + if nesterov: + update = g.add(buf, alpha=momentum) + else: + update = buf + + update = zeropower_via_newtonschulz5(update, steps=backend_steps) + + if sharded: + prev_ag_handle = dist.all_gather_into_tensor( + m['full_update'], update, async_op=True) + prev_m = m + else: + if wd > 0.0: + p.data.mul_(1.0 - lr * wd) + p.add_(update.to(dtype=p.dtype), alpha=-lr * m['scale']) + + if prev_ag_handle is not None: + prev_ag_handle.wait() + pp = prev_m['p'] + upd = prev_m['full_update'][:prev_m['B']] + if wd > 0.0: + pp.data.mul_(1.0 - lr * wd) + pp.add_(upd.to(dtype=pp.dtype), alpha=-lr * prev_m['scale']) + + if hasattr(self, '_rs_futures'): + del self._rs_futures + + return loss + +# --- Tokenizer evaluation helpers --- + +def build_sentencepiece_luts( + sp: spm.SentencePieceProcessor, vocab_size: int, device: torch.device +) -> tuple[Tensor, Tensor, Tensor]: + sp_vocab_size = int(sp.vocab_size()) + table_size = max(sp_vocab_size, vocab_size) + base_bytes_np = np.zeros((table_size,), dtype=np.int16) + has_leading_space_np = np.zeros((table_size,), dtype=np.bool_) + is_boundary_token_np = np.ones((table_size,), dtype=np.bool_) + for token_id in range(sp_vocab_size): + if sp.is_control(token_id) or sp.is_unknown(token_id) or sp.is_unused(token_id): + continue + is_boundary_token_np[token_id] = False + if sp.is_byte(token_id): + base_bytes_np[token_id] = 1 + continue + piece = sp.id_to_piece(token_id) + if piece.startswith("\u2581"): + has_leading_space_np[token_id] = True + piece = piece[1:] + base_bytes_np[token_id] = len(piece.encode("utf-8")) + return ( + torch.tensor(base_bytes_np, dtype=torch.int16, device=device), + torch.tensor(has_leading_space_np, dtype=torch.bool, device=device), + torch.tensor(is_boundary_token_np, dtype=torch.bool, device=device), + ) +def load_validation_tokens(pattern: str, seq_len: int) -> Tensor: + files = [Path(p) for p in sorted(glob.glob(pattern))] + if not files: + raise FileNotFoundError(f"No files found for pattern: {pattern}") + tokens = torch.cat([load_data_shard(file) for file in files]).contiguous() + usable = ((tokens.numel() - 1) // seq_len) * seq_len + if usable <= 0: + raise ValueError(f"Validation split is too short for TRAIN_SEQ_LEN={seq_len}") + return tokens[: usable + 1] +def eval_val( + args: Hyperparameters, + model: nn.Module, + rank: int, + world_size: int, + device: torch.device, + grad_accum_steps: int, + val_tokens: Tensor, + base_bytes_lut: Tensor, + has_leading_space_lut: Tensor, + is_boundary_token_lut: Tensor, + eval_seq_len: int | None = None, +) -> tuple[float, float]: + seq_len = eval_seq_len or args.train_seq_len + local_batch_tokens = args.val_batch_size // (world_size * grad_accum_steps) + if local_batch_tokens < seq_len: + raise ValueError( + "VAL_BATCH_SIZE must provide at least one sequence per rank; " + f"got VAL_BATCH_SIZE={args.val_batch_size}, WORLD_SIZE={world_size}, " + f"GRAD_ACCUM_STEPS={grad_accum_steps}, seq_len={seq_len}" + ) + local_batch_seqs = local_batch_tokens // seq_len + total_seqs = (val_tokens.numel() - 1) // seq_len + seq_start = (total_seqs * rank) // world_size + seq_end = (total_seqs * (rank + 1)) // world_size + val_loss_sum = torch.zeros((), device=device, dtype=torch.float64) + val_token_count = torch.zeros((), device=device, dtype=torch.float64) + val_byte_count = torch.zeros((), device=device, dtype=torch.float64) + model.eval() + with torch.inference_mode(): + for batch_seq_start in range(seq_start, seq_end, local_batch_seqs): + batch_seq_end = min(batch_seq_start + local_batch_seqs, seq_end) + raw_start = batch_seq_start * seq_len + raw_end = batch_seq_end * seq_len + 1 + local = val_tokens[raw_start:raw_end].to(device=device, dtype=torch.int64, non_blocking=True) + x = local[:-1].reshape(-1, seq_len) + y = local[1:].reshape(-1, seq_len) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + batch_loss = model(x, y).detach() + batch_token_count = float(y.numel()) + val_loss_sum += batch_loss.to(torch.float64) * batch_token_count + val_token_count += batch_token_count + prev_ids = x.reshape(-1) + tgt_ids = y.reshape(-1) + token_bytes = base_bytes_lut[tgt_ids].to(dtype=torch.int16) + token_bytes += (has_leading_space_lut[tgt_ids] & ~is_boundary_token_lut[prev_ids]).to(dtype=torch.int16) + val_byte_count += token_bytes.to(torch.float64).sum() + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(val_loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(val_token_count, op=dist.ReduceOp.SUM) + dist.all_reduce(val_byte_count, op=dist.ReduceOp.SUM) + val_loss = val_loss_sum / val_token_count + bits_per_token = val_loss.item() / math.log(2.0) + tokens_per_byte = val_token_count.item() / val_byte_count.item() + model.train() + return float(val_loss.item()), float(bits_per_token * tokens_per_byte) + +# --- Quantization helpers --- + +CONTROL_TENSOR_NAME_PATTERNS = tuple( + pattern + for pattern in os.environ.get( + "CONTROL_TENSOR_NAME_PATTERNS", + "attn_scale,attn_scales,mlp_scale,mlp_scales,resid_mix,resid_mixes,q_gain,skip_weight,skip_weights,skip_gates,smear,dtg_gate,ve_layer_scales,ve_shared.scale,attn_gate,vr_lambda", + ).split(",") + if pattern +) +INT8_KEEP_FLOAT_FP32_NAME_PATTERNS = tuple( + pattern + for pattern in os.environ.get( + "INT8_KEEP_FLOAT_FP32_NAME_PATTERNS", + ",".join(CONTROL_TENSOR_NAME_PATTERNS), + ).split(",") + if pattern +) +INT8_KEEP_FLOAT_MAX_NUMEL = 65_536 +INT8_KEEP_FLOAT_STORE_DTYPE = torch.float16 +INT8_PER_ROW_SCALE_DTYPE = torch.float16 +INT8_CLIP_PERCENTILE = 99.99984 +INT8_CLIP_Q = INT8_CLIP_PERCENTILE / 100.0 +def tensor_nbytes(t: Tensor) -> int: + return int(t.numel()) * int(t.element_size()) +def keep_float_tensor(name: str, t: Tensor, passthrough_orig_dtypes: dict[str, str]) -> Tensor: + if any(pattern in name for pattern in INT8_KEEP_FLOAT_FP32_NAME_PATTERNS): + return t.float().contiguous() + if t.dtype in {torch.float32, torch.bfloat16}: + passthrough_orig_dtypes[name] = str(t.dtype).removeprefix("torch.") + return t.to(dtype=INT8_KEEP_FLOAT_STORE_DTYPE).contiguous() + return t +def quantize_float_tensor(t: Tensor) -> tuple[Tensor, Tensor]: + t32 = t.float() + if t32.ndim == 2: + clip_abs = ( + torch.quantile(t32.abs(), INT8_CLIP_Q, dim=1) + if t32.numel() + else torch.empty((t32.shape[0],), dtype=torch.float32) + ) + clipped = torch.maximum(torch.minimum(t32, clip_abs[:, None]), -clip_abs[:, None]) + scale = (clip_abs / 127.0).clamp_min(1.0 / 127.0) + q = torch.clamp(torch.round(clipped / scale[:, None]), -127, 127).to(torch.int8).contiguous() + return q, scale.to(dtype=INT8_PER_ROW_SCALE_DTYPE).contiguous() + clip_abs = float(torch.quantile(t32.abs().flatten(), INT8_CLIP_Q).item()) if t32.numel() else 0.0 + scale = torch.tensor(clip_abs / 127.0 if clip_abs > 0 else 1.0, dtype=torch.float32) + q = torch.clamp(torch.round(torch.clamp(t32, -clip_abs, clip_abs) / scale), -127, 127).to(torch.int8).contiguous() + return q, scale +def quantize_state_dict_int8(state_dict: dict[str, Tensor]): + quantized: dict[str, Tensor] = {} + scales: dict[str, Tensor] = {} + dtypes: dict[str, str] = {} + passthrough: dict[str, Tensor] = {} + passthrough_orig_dtypes: dict[str, str] = {} + qmeta: dict[str, dict[str, object]] = {} + stats = dict.fromkeys( + ("param_count", "num_tensors", "num_float_tensors", "num_nonfloat_tensors", "baseline_tensor_bytes", "int8_payload_bytes"), + 0, + ) + for name, tensor in state_dict.items(): + t = tensor.detach().to("cpu").contiguous() + stats["param_count"] += int(t.numel()) + stats["num_tensors"] += 1 + stats["baseline_tensor_bytes"] += tensor_nbytes(t) + if not t.is_floating_point(): + stats["num_nonfloat_tensors"] += 1 + passthrough[name] = t + stats["int8_payload_bytes"] += tensor_nbytes(t) + continue + if t.numel() <= INT8_KEEP_FLOAT_MAX_NUMEL: + kept = keep_float_tensor(name, t, passthrough_orig_dtypes) + passthrough[name] = kept + stats["int8_payload_bytes"] += tensor_nbytes(kept) + continue + stats["num_float_tensors"] += 1 + q, s = quantize_float_tensor(t) + if s.ndim > 0: + qmeta[name] = {"scheme": "per_row", "axis": 0} + quantized[name] = q + scales[name] = s + dtypes[name] = str(t.dtype).removeprefix("torch.") + stats["int8_payload_bytes"] += tensor_nbytes(q) + tensor_nbytes(s) + obj: dict[str, object] = { + "__quant_format__": "int8_clean_per_row_v1", + "quantized": quantized, + "scales": scales, + "dtypes": dtypes, + "passthrough": passthrough, + } + if qmeta: + obj["qmeta"] = qmeta + if passthrough_orig_dtypes: + obj["passthrough_orig_dtypes"] = passthrough_orig_dtypes + return obj, stats +def dequantize_state_dict_int8(obj: dict[str, object]) -> dict[str, Tensor]: + out: dict[str, Tensor] = {} + qmeta = obj.get("qmeta", {}) + passthrough_orig_dtypes = obj.get("passthrough_orig_dtypes", {}) + for name, q in obj["quantized"].items(): + dtype = getattr(torch, obj["dtypes"][name]) + s = obj["scales"][name] + if qmeta.get(name, {}).get("scheme") == "per_row" or s.ndim > 0: + s = s.to(dtype=torch.float32) + out[name] = (q.float() * s.view(q.shape[0], *([1] * (q.ndim - 1)))).to(dtype=dtype).contiguous() + else: + scale = float(s.item()) + out[name] = (q.float() * scale).to(dtype=dtype).contiguous() + for name, t in obj["passthrough"].items(): + out_t = t.detach().to("cpu").contiguous() + orig_dtype = passthrough_orig_dtypes.get(name) + if isinstance(orig_dtype, str): + out_t = out_t.to(dtype=getattr(torch, orig_dtype)).contiguous() + out[name] = out_t + return out + +# --- Data loading --- + +def load_data_shard(file: Path) -> Tensor: + header_bytes = 256 * np.dtype(" None: + self.file_idx = (self.file_idx + 1) % len(self.files) + self.tokens = load_data_shard(self.files[self.file_idx]) + self.pos = 0 + def take(self, n: int) -> Tensor: + chunks: list[Tensor] = [] + remaining = n + while remaining > 0: + avail = self.tokens.numel() - self.pos + if avail <= 0: + self._advance_file() + continue + k = min(remaining, avail) + chunks.append(self.tokens[self.pos : self.pos + k]) + self.pos += k + remaining -= k + return chunks[0] if len(chunks) == 1 else torch.cat(chunks) +class DistributedTokenLoader: + def __init__(self, pattern: str, rank: int, world_size: int, device: torch.device): + self.rank = rank + self.world_size = world_size + self.device = device + self.stream = TokenStream(pattern) + def next_batch(self, global_tokens: int, seq_len: int, grad_accum_steps: int) -> tuple[Tensor, Tensor]: + local_tokens = global_tokens // (self.world_size * grad_accum_steps) + per_rank_span = local_tokens + 1 + chunk = self.stream.take(per_rank_span * self.world_size) + start = self.rank * per_rank_span + local = chunk[start : start + per_rank_span].to(dtype=torch.int64) + x = local[:-1].reshape(-1, seq_len) + y = local[1:].reshape(-1, seq_len) + return x.to(self.device, non_blocking=True), y.to(self.device, non_blocking=True) + +# --- Transformer modules --- + +class RMSNorm(nn.Module): + def __init__(self, eps: float | None = None): + super().__init__() + self.eps = eps + def forward(self, x: Tensor) -> Tensor: + return F.rms_norm(x, (x.size(-1),), eps=self.eps) +class CastedLinear(nn.Linear): + _qat_enabled: bool = False + def forward(self, x: Tensor) -> Tensor: + w = self.weight.to(x.dtype) + if CastedLinear._qat_enabled and self.training and w.ndim == 2: + with torch.no_grad(): + w32 = self.weight.float() + row_max = w32.abs().amax(dim=1) + scale = (row_max / 31.0).clamp_min(1.0 / 31.0) + w_q = (torch.clamp(torch.round(w32 / scale[:, None]), -32, 31) * scale[:, None]).to(x.dtype) + w = w + (w_q - w).detach() + bias = self.bias.to(x.dtype) if self.bias is not None else None + return F.linear(x, w, bias) +def restore_low_dim_params_to_fp32(module: nn.Module) -> None: + with torch.no_grad(): + for name, param in module.named_parameters(): + if (param.ndim < 2 or any(pattern in name for pattern in CONTROL_TENSOR_NAME_PATTERNS)) and param.dtype != torch.float32: + param.data = param.data.float() +class Rotary(nn.Module): + def __init__(self, dim: int, base: float = 10000.0, train_seq_len: int = 1024, rope_dims: int = 0): + super().__init__() + self.dim = dim + self.base = base + self.train_seq_len = train_seq_len + self.rope_dims = rope_dims if rope_dims > 0 else dim + inv_freq = 1.0 / (base ** (torch.arange(0, self.rope_dims, 2, dtype=torch.float32) / self.rope_dims)) + self.register_buffer("inv_freq", inv_freq, persistent=False) + self._seq_len_cached = 0 + self._cos_cached: Tensor | None = None + self._sin_cached: Tensor | None = None + def forward(self, seq_len: int, device: torch.device, dtype: torch.dtype) -> tuple[Tensor, Tensor]: + if ( + self._cos_cached is None + or self._sin_cached is None + or self._seq_len_cached != seq_len + or self._cos_cached.device != device + ): + rd = self.rope_dims + if seq_len > self.train_seq_len: + scale = seq_len / self.train_seq_len + new_base = self.base * (scale ** (rd / (rd - 2))) + inv_freq = 1.0 / (new_base ** (torch.arange(0, rd, 2, dtype=torch.float32, device=device) / rd)) + else: + inv_freq = self.inv_freq.to(device) + t = torch.arange(seq_len, device=device, dtype=inv_freq.dtype) + freqs = torch.outer(t, inv_freq) + self._cos_cached = freqs.cos()[None, :, None, :] + self._sin_cached = freqs.sin()[None, :, None, :] + self._seq_len_cached = seq_len + return self._cos_cached.to(dtype=dtype), self._sin_cached.to(dtype=dtype) +def apply_rotary_emb(x: Tensor, cos: Tensor, sin: Tensor, rope_dims: int = 0) -> Tensor: + if rope_dims > 0 and rope_dims < x.size(-1): + x_rope, x_pass = x[..., :rope_dims], x[..., rope_dims:] + half = rope_dims // 2 + x1, x2 = x_rope[..., :half], x_rope[..., half:] + x_rope = torch.cat((x1 * cos + x2 * sin, x1 * (-sin) + x2 * cos), dim=-1) + return torch.cat((x_rope, x_pass), dim=-1) + half = x.size(-1) // 2 + x1, x2 = x[..., :half], x[..., half:] + return torch.cat((x1 * cos + x2 * sin, x1 * (-sin) + x2 * cos), dim=-1) + +class CausalSelfAttention(nn.Module): + def __init__( + self, + dim: int, + num_heads: int, + num_kv_heads: int, + rope_base: float, + qk_gain_init: float, + gated_attention: bool = False, + value_residual: bool = False, + ): + super().__init__() + if dim % num_heads != 0: + raise ValueError("model_dim must be divisible by num_heads") + if num_heads % num_kv_heads != 0: + raise ValueError("num_heads must be divisible by num_kv_heads") + self.num_heads = num_heads + self.num_kv_heads = num_kv_heads + self.head_dim = dim // num_heads + if self.head_dim % 2 != 0: + raise ValueError("head_dim must be even for RoPE") + # No CastedLinear -- weights come from banks + self.q_gain = nn.Parameter(torch.full((num_heads,), qk_gain_init, dtype=torch.float32)) + self.rope_dims = 0 # set by GPT.__init__ for partial RoPE + self.rotary = Rotary(self.head_dim, base=rope_base, train_seq_len=1024) + self.use_xsa = False # set by GPT.__init__ for deep layers only + # Gated attention and value residual (non-banked small params) + self.gated_attention = gated_attention + if gated_attention: + self.attn_gate = nn.Linear(dim, num_heads, bias=True) + nn.init.zeros_(self.attn_gate.weight) + nn.init.constant_(self.attn_gate.bias, 4.0) + self.value_residual = value_residual + if value_residual: + self.vrl_alpha = nn.Parameter(torch.zeros(1, dtype=torch.float32)) # sigmoid gate (PR #569 style) + def _xsa_efficient(self, y: Tensor, v: Tensor) -> Tensor: + """Efficient XSA: subtract self-value projection via GQA-aware reshape (no repeat_interleave). + y: [B, T, H, D], v: [B, T, Hkv, D]. H must be divisible by Hkv.""" + B, T, H, D = y.shape + Hkv = v.size(-2) + group = H // Hkv + y_g = y.reshape(B, T, Hkv, group, D) # [B, T, Hkv, group, D] + vn = F.normalize(v, dim=-1).unsqueeze(-2) # [B, T, Hkv, 1, D] -- broadcast ready + proj = (y_g * vn).sum(dim=-1, keepdim=True) * vn + return (y_g - proj).reshape(B, T, H, D) + def forward(self, x: Tensor, q_w: Tensor, k_w: Tensor, v_w: Tensor, out_w: Tensor, v_embed: Tensor | None = None, v0: Tensor | None = None) -> tuple[Tensor, Tensor | None]: + bsz, seqlen, dim = x.shape + q = F.linear(x, q_w.to(x.dtype)).reshape(bsz, seqlen, self.num_heads, self.head_dim) + k = F.linear(x, k_w.to(x.dtype)).reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + v = F.linear(x, v_w.to(x.dtype)) + if v_embed is not None: + v = v + v_embed + v = v.reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + raw_v = v if self.value_residual else None + if self.value_residual and v0 is not None: + alpha = torch.sigmoid(self.vrl_alpha.to(dtype=v.dtype)) + v = v + alpha * v0 # sigmoid-gated residual (PR #569 style) + q = F.rms_norm(q, (q.size(-1),)) + k = F.rms_norm(k, (k.size(-1),)) + cos, sin = self.rotary(seqlen, x.device, q.dtype) + q = apply_rotary_emb(q, cos, sin, self.rope_dims) + k = apply_rotary_emb(k, cos, sin, self.rope_dims) + q = q * self.q_gain.to(dtype=q.dtype)[None, None, :, None] + if HAS_FA3: + y = flash_attn_3_func(q, k, v, causal=True) + else: + q2, k2, v2 = q.transpose(1, 2), k.transpose(1, 2), v.transpose(1, 2) + if k2.size(1) != q2.size(1): + reps = q2.size(1) // k2.size(1) + k2 = k2.repeat_interleave(reps, dim=1) + v2 = v2.repeat_interleave(reps, dim=1) + y = F.scaled_dot_product_attention(q2, k2, v2, is_causal=True).transpose(1, 2) + if self.use_xsa: + y = self._xsa_efficient(y, v) + if self.gated_attention: + # gate shape: (bsz, seqlen, num_heads) -> (bsz, seqlen, num_heads, 1) for B,T,H,D layout + gate = torch.sigmoid(self.attn_gate(x)).unsqueeze(-1) + y = y * gate + y = y.reshape(bsz, seqlen, dim) + return F.linear(y, out_w.to(x.dtype)), raw_v + +class SmearGate(nn.Module): + def __init__(self, dim: int): + super().__init__() + self.gate = nn.Parameter(torch.zeros(dim, dtype=torch.float32)) + def forward(self, x: Tensor) -> Tensor: + g = torch.sigmoid(self.gate.to(dtype=x.dtype))[None, None, :] + x_prev = torch.cat([torch.zeros_like(x[:, :1]), x[:, :-1]], dim=1) + return (1 - g) * x + g * x_prev + +class BigramHashEmbedding(nn.Module): + def __init__(self, bigram_vocab_size: int, bigram_dim: int, model_dim: int, trigram: bool = False): + super().__init__() + self.bigram_vocab_size = bigram_vocab_size + self._trigram = trigram + self.embed = nn.Embedding(bigram_vocab_size, bigram_dim) + nn.init.zeros_(self.embed.weight) + self.proj = CastedLinear(bigram_dim, model_dim, bias=False) if bigram_dim != model_dim else None + if self.proj is not None: + nn.init.zeros_(self.proj.weight) + self.scale = nn.Parameter(torch.tensor(0.05, dtype=torch.float32)) + def bigram_hash(self, tokens: Tensor) -> Tensor: + t = tokens.to(torch.int32) + mod = self.bigram_vocab_size - 1 + out = torch.empty_like(t) + out[..., 0] = mod + out[..., 1:] = torch.bitwise_xor(36313 * t[..., 1:], 27191 * t[..., :-1]) % mod + return out.long() + def trigram_hash(self, tokens: Tensor) -> Tensor: + """Hash (t-2, t-1, t) trigrams into same embedding table. Zero extra params.""" + t = tokens.to(torch.int32) + mod = self.bigram_vocab_size - 1 + out = torch.empty_like(t) + out[..., :2] = mod + out[..., 2:] = (36313 * t[..., 2:] ^ 27191 * t[..., 1:-1] ^ 51497 * t[..., :-2]) % mod + return out.long() + def forward(self, token_ids: Tensor) -> Tensor: + h = self.embed(self.bigram_hash(token_ids)) + if self._trigram: + h = h + self.embed(self.trigram_hash(token_ids)) + if self.proj is not None: + h = self.proj(h) + return h * self.scale.to(dtype=h.dtype) + +class ValueEmbedding(nn.Module): + """Reinject token identity into attention values at specific layers. + Each table maps vocab tokens to a low-dim embedding, projected to model_dim.""" + def __init__(self, vocab_size: int, ve_dim: int, model_dim: int): + super().__init__() + self.embed = nn.Embedding(vocab_size, ve_dim) + nn.init.normal_(self.embed.weight, std=0.01) + self.proj = CastedLinear(ve_dim, model_dim, bias=False) if ve_dim != model_dim else None + if self.proj is not None: + nn.init.zeros_(self.proj.weight) + self.scale = nn.Parameter(torch.tensor(0.1, dtype=torch.float32)) + def forward(self, token_ids: Tensor) -> Tensor: + h = self.embed(token_ids) + if self.proj is not None: + h = self.proj(h) + return h * self.scale.to(dtype=h.dtype) + +class MLP(nn.Module): + def __init__(self, dim: int, mlp_mult: int): + super().__init__() + # No CastedLinear -- weights come from banks + def forward(self, x: Tensor, up_w: Tensor, down_w: Tensor) -> Tensor: + x = F.leaky_relu(F.linear(x, up_w.to(x.dtype)), negative_slope=0.5) + return F.linear(x.square(), down_w.to(x.dtype)) + +class Block(nn.Module): + def __init__( + self, + dim: int, + num_heads: int, + num_kv_heads: int, + mlp_mult: int, + rope_base: float, + qk_gain_init: float, + layer_idx: int = 0, + ln_scale: bool = False, + dtg: bool = False, + gated_attention: bool = False, + value_residual: bool = False, + ): + super().__init__() + self.attn_norm = RMSNorm() + self.mlp_norm = RMSNorm() + self.attn = CausalSelfAttention(dim, num_heads, num_kv_heads, rope_base, qk_gain_init, + gated_attention=gated_attention, value_residual=value_residual) + self.mlp = MLP(dim, mlp_mult) + self.attn_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.mlp_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.resid_mix = nn.Parameter(torch.stack((torch.ones(dim), torch.zeros(dim))).float()) + self.ln_scale_factor = 1.0 / math.sqrt(layer_idx + 1) if ln_scale else 1.0 + if dtg: + self.dtg_gate = nn.Linear(dim, 1, bias=True) + nn.init.zeros_(self.dtg_gate.weight) + nn.init.constant_(self.dtg_gate.bias, 2.0) + else: + self.dtg_gate = None + def forward(self, x: Tensor, x0: Tensor, q_w: Tensor, k_w: Tensor, v_w: Tensor, out_w: Tensor, up_w: Tensor, down_w: Tensor, v_embed: Tensor | None = None, v0: Tensor | None = None) -> tuple[Tensor, Tensor | None]: + mix = self.resid_mix.to(dtype=x.dtype) + x_in = mix[0][None, None, :] * x + mix[1][None, None, :] * x0 + attn_out, raw_v = self.attn(self.attn_norm(x_in) * self.ln_scale_factor, q_w, k_w, v_w, out_w, v_embed=v_embed, v0=v0) + x_out = x_in + self.attn_scale.to(dtype=x_in.dtype)[None, None, :] * attn_out + x_out = x_out + self.mlp_scale.to(dtype=x_out.dtype)[None, None, :] * self.mlp(self.mlp_norm(x_out) * self.ln_scale_factor, up_w, down_w) + if self.dtg_gate is not None: + gate = torch.sigmoid(self.dtg_gate(x_in.detach())) + x_out = x_in + gate * (x_out - x_in) + return x_out, raw_v + +class GPT(nn.Module): + def __init__( + self, + vocab_size: int, + num_layers: int, + model_dim: int, + num_heads: int, + num_kv_heads: int, + mlp_mult: int, + tie_embeddings: bool, + tied_embed_init_std: float, + logit_softcap: float, + rope_base: float, + qk_gain_init: float, + mtp_num_heads: int = 0, + mtp_loss_weight: float = 0.1, + bigram_vocab_size: int = 0, + bigram_dim: int = 128, + xsa_last_n: int = 0, + rope_dims: int = 0, + ln_scale: bool = False, + dtg: bool = False, + ve_enabled: bool = False, + ve_dim: int = 128, + ve_layers: str = "9,10", + gated_attention: bool = False, + value_residual: bool = False, + skip_gates_enabled: bool = True, + ): + super().__init__() + self._ve_target_dim = num_kv_heads * (model_dim // num_heads) # kv_dim for value projection + if logit_softcap <= 0.0: + raise ValueError(f"logit_softcap must be positive, got {logit_softcap}") + self.tie_embeddings = tie_embeddings + self.tied_embed_init_std = tied_embed_init_std + self.logit_softcap = logit_softcap + self.value_residual = value_residual + self.mtp_num_heads = mtp_num_heads + self.mtp_loss_weight = mtp_loss_weight + self.tok_emb = nn.Embedding(vocab_size, model_dim) + self.bigram = BigramHashEmbedding(bigram_vocab_size, bigram_dim, model_dim, trigram=bool(int(os.environ.get("TRIGRAM", "0")))) if bigram_vocab_size > 0 else None + self.smear = SmearGate(model_dim) + self.num_encoder_layers = num_layers // 2 + self.num_decoder_layers = num_layers - self.num_encoder_layers + self.num_skip_weights = min(self.num_encoder_layers, self.num_decoder_layers) + self.skip_weights = nn.Parameter(torch.ones(self.num_skip_weights, model_dim, dtype=torch.float32)) + self.skip_gates = nn.Parameter(torch.zeros(self.num_skip_weights, model_dim, dtype=torch.float32)) if skip_gates_enabled else None + # Parameter banks: contiguous 3D tensors for batched optimizer + head_dim = model_dim // num_heads + kv_dim = num_kv_heads * head_dim + mlp_dim = int(mlp_mult * model_dim) + self.num_layers = num_layers + self.qo_bank = nn.Parameter(torch.empty(2 * num_layers, model_dim, model_dim)) + self.kv_bank = nn.Parameter(torch.empty(2 * num_layers, kv_dim, model_dim)) + self.mlp_up_bank = nn.Parameter(torch.empty(num_layers, mlp_dim, model_dim)) + self.mlp_down_bank = nn.Parameter(torch.empty(num_layers, model_dim, mlp_dim)) + self.blocks = nn.ModuleList( + [ + Block( + model_dim, + num_heads, + num_kv_heads, + mlp_mult, + rope_base, + qk_gain_init, + layer_idx=i, + ln_scale=ln_scale, + dtg=dtg, + gated_attention=gated_attention, + value_residual=value_residual, + ) + for i in range(num_layers) + ] + ) + if rope_dims > 0: + head_dim = model_dim // num_heads + for block in self.blocks: + block.attn.rope_dims = rope_dims + block.attn.rotary = Rotary(head_dim, base=rope_base, train_seq_len=1024, rope_dims=rope_dims) + self.ve_layer_indices = [int(x) for x in ve_layers.split(",") if x.strip()] if ve_enabled else [] + kv_dim_ve = self._ve_target_dim + if self.ve_layer_indices: + self.ve_shared = ValueEmbedding(vocab_size, ve_dim, kv_dim_ve) + self.ve_layer_scales = nn.ParameterList( + [nn.Parameter(torch.ones(1, dtype=torch.float32)) for _ in self.ve_layer_indices] + ) + else: + self.ve_shared = None + self.ve_layer_scales = nn.ParameterList() + self.value_embeds = nn.ModuleList() # keep empty for compat + self.final_norm = RMSNorm() + self.lm_head = None if tie_embeddings else CastedLinear(model_dim, vocab_size, bias=False) + if self.lm_head is not None: + self.lm_head._zero_init = True + self.mtp_heads = nn.ModuleList( + [CastedLinear(model_dim, vocab_size, bias=False) for _ in range(mtp_num_heads)] + ) + for head in self.mtp_heads: + head._zero_init = True + if xsa_last_n > 0: + for i in range(max(0, num_layers - xsa_last_n), num_layers): + self.blocks[i].attn.use_xsa = True + self._init_weights() + def _init_weights(self) -> None: + if self.tie_embeddings: + nn.init.normal_(self.tok_emb.weight, mean=0.0, std=self.tied_embed_init_std) + n = self.num_layers + proj_scale = 1.0 / math.sqrt(2 * n) + # Init banks: orthogonal, with proj layers scaled down and out/down zero-init + for i in range(n): + nn.init.orthogonal_(self.qo_bank.data[i], gain=1.0) # Q + nn.init.zeros_(self.qo_bank.data[n + i]) # Out (zero init) + nn.init.orthogonal_(self.kv_bank.data[i], gain=1.0) # K + nn.init.orthogonal_(self.kv_bank.data[n + i], gain=1.0) # V + nn.init.orthogonal_(self.mlp_up_bank.data[i], gain=1.0) # MLP up + nn.init.zeros_(self.mlp_down_bank.data[i]) # MLP down (zero init) + # Scale proj layers (out_proj and mlp_down are "proj" layers) + self.qo_bank.data[n + i].mul_(proj_scale) + self.mlp_down_bank.data[i].mul_(proj_scale) + # Init remaining nn.Linear modules (bigram proj, mtp heads, lm_head) + for name, module in self.named_modules(): + if isinstance(module, nn.Linear): + if getattr(module, "_zero_init", False): + nn.init.zeros_(module.weight) + elif module.weight.ndim == 2 and module.weight.shape[0] >= 64 and module.weight.shape[1] >= 64: + nn.init.orthogonal_(module.weight, gain=1.0) + def _get_ve(self, layer_idx: int, input_ids: Tensor, ve_cache: dict | None = None) -> Tensor | None: + """Get value embedding for a specific layer using shared table + per-layer scale.""" + if self.ve_shared is None or layer_idx not in self.ve_layer_indices: + return None + if ve_cache is not None and 've' not in ve_cache: + ve_cache['ve'] = self.ve_shared(input_ids) + ve_base = ve_cache['ve'] if ve_cache is not None else self.ve_shared(input_ids) + ve_idx = self.ve_layer_indices.index(layer_idx) + return ve_base * self.ve_layer_scales[ve_idx].to(dtype=ve_base.dtype) + def forward(self, input_ids: Tensor, target_ids: Tensor) -> Tensor: + n = self.num_layers + x = self.tok_emb(input_ids) + if self.bigram is not None: + x = x + self.bigram(input_ids) + x = F.rms_norm(x, (x.size(-1),)) + x = self.smear(x) + x0 = x + v0 = None + skips: list[Tensor] = [] + ve_cache: dict = {} + for i in range(self.num_encoder_layers): + ve = self._get_ve(i, input_ids, ve_cache) + x, raw_v = self.blocks[i](x, x0, + self.qo_bank[i], self.kv_bank[i], self.kv_bank[n + i], + self.qo_bank[n + i], self.mlp_up_bank[i], self.mlp_down_bank[i], + v_embed=ve, v0=v0) + if v0 is None and raw_v is not None: + v0 = raw_v + skips.append(x) + for i in range(self.num_decoder_layers): + bi = self.num_encoder_layers + i + if skips: + scaled_skip = self.skip_weights[i].to(dtype=x.dtype)[None, None, :] * skips.pop() + if self.skip_gates is not None: + g = torch.sigmoid(self.skip_gates[i].to(dtype=x.dtype))[None, None, :] + x = torch.lerp(scaled_skip, x, g) + else: + x = x + scaled_skip + ve = self._get_ve(bi, input_ids, ve_cache) + x, _ = self.blocks[bi](x, x0, + self.qo_bank[bi], self.kv_bank[bi], self.kv_bank[n + bi], + self.qo_bank[n + bi], self.mlp_up_bank[bi], self.mlp_down_bank[bi], + v_embed=ve, v0=v0) + x = self.final_norm(x) + x_flat = x.reshape(-1, x.size(-1)) + targets = target_ids.reshape(-1) + if self.tie_embeddings: + logits_proj = F.linear(x_flat, self.tok_emb.weight) + else: + if self.lm_head is None: + raise RuntimeError("lm_head is required when tie_embeddings=False") + logits_proj = self.lm_head(x_flat) + logits = self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + main_loss = F.cross_entropy(logits.float(), targets, reduction="mean") + if self.training and self.mtp_num_heads > 0 and self.mtp_loss_weight > 0.0: + _, seqlen, dim = x.shape + mtp_loss_sum = x.new_zeros(()) + mtp_loss_count = 0 + for k, mtp_head in enumerate(self.mtp_heads): + valid_t = seqlen - (k + 1) + if valid_t <= 0: + continue + mtp_hidden = x[:, :valid_t, :].reshape(-1, dim) + mtp_targets = target_ids[:, k + 1 :].reshape(-1) + mtp_logits_proj = mtp_head(mtp_hidden) + mtp_logits = self.logit_softcap * torch.tanh(mtp_logits_proj / self.logit_softcap) + mtp_loss_sum = mtp_loss_sum + F.cross_entropy(mtp_logits.float(), mtp_targets, reduction="mean") + mtp_loss_count += 1 + if mtp_loss_count > 0: + main_loss = main_loss + self.mtp_loss_weight * (mtp_loss_sum / mtp_loss_count) + return main_loss + def forward_hidden(self, input_ids: Tensor) -> Tensor: + """Return final hidden states before lm_head.""" + n = self.num_layers + x = self.tok_emb(input_ids) + if self.bigram is not None: + x = x + self.bigram(input_ids) + x = F.rms_norm(x, (x.size(-1),)) + x = self.smear(x) + x0 = x + v0 = None + skips: list[Tensor] = [] + ve_cache: dict = {} + for i in range(self.num_encoder_layers): + ve = self._get_ve(i, input_ids, ve_cache) + x, raw_v = self.blocks[i](x, x0, + self.qo_bank[i], self.kv_bank[i], self.kv_bank[n + i], + self.qo_bank[n + i], self.mlp_up_bank[i], self.mlp_down_bank[i], + v_embed=ve, v0=v0) + if v0 is None and raw_v is not None: + v0 = raw_v + skips.append(x) + for i in range(self.num_decoder_layers): + bi = self.num_encoder_layers + i + if skips: + scaled_skip = self.skip_weights[i].to(dtype=x.dtype)[None, None, :] * skips.pop() + if self.skip_gates is not None: + g = torch.sigmoid(self.skip_gates[i].to(dtype=x.dtype))[None, None, :] + x = torch.lerp(scaled_skip, x, g) + else: + x = x + scaled_skip + ve = self._get_ve(bi, input_ids, ve_cache) + x, _ = self.blocks[bi](x, x0, + self.qo_bank[bi], self.kv_bank[bi], self.kv_bank[n + bi], + self.qo_bank[n + bi], self.mlp_up_bank[bi], self.mlp_down_bank[bi], + v_embed=ve, v0=v0) + return self.final_norm(x) + + def compute_logits(self, hidden_states: Tensor) -> Tensor: + """Compute logits from hidden states.""" + if self.tie_embeddings: + logits_proj = F.linear(hidden_states, self.tok_emb.weight) + else: + logits_proj = self.lm_head(hidden_states) + return self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + + def forward_logits(self, input_ids: Tensor) -> Tensor: + """Return logits (bsz, seq_len, vocab) without computing loss.""" + return self.compute_logits(self.forward_hidden(input_ids)) + +# --- Sliding window evaluation --- + +def eval_val_sliding( + args: Hyperparameters, + base_model: nn.Module, + rank: int, + world_size: int, + device: torch.device, + val_tokens: Tensor, + base_bytes_lut: Tensor, + has_leading_space_lut: Tensor, + is_boundary_token_lut: Tensor, + stride: int, + batch_seqs: int = 32, + eval_seq_len: int | None = None, +) -> tuple[float, float]: + """Sliding window evaluation: each token scored with maximum context.""" + seq_len = eval_seq_len or args.train_seq_len + total_tokens = val_tokens.numel() - 1 + window_starts = [ws for ws in range(0, total_tokens, stride) + if min(ws + seq_len, total_tokens) - ws >= 1] + total_windows = len(window_starts) + my_s = (total_windows * rank) // world_size + my_e = (total_windows * (rank + 1)) // world_size + my_windows = window_starts[my_s:my_e] + loss_sum = torch.zeros((), device=device, dtype=torch.float64) + token_count = torch.zeros((), device=device, dtype=torch.float64) + byte_count = torch.zeros((), device=device, dtype=torch.float64) + base_model.eval() + compiled_logits = torch.compile(base_model.forward_logits, dynamic=False, fullgraph=True) + with torch.inference_mode(): + for bi in range(0, len(my_windows), batch_seqs): + batch_ws = my_windows[bi:bi + batch_seqs] + bsz = len(batch_ws) + x_batch = torch.zeros(bsz, seq_len, dtype=torch.int64, device=device) + y_batch = torch.zeros(bsz, seq_len, dtype=torch.int64, device=device) + wlens: list[int] = [] + for i, ws in enumerate(batch_ws): + end = min(ws + seq_len, total_tokens) + wlen = end - ws + wlens.append(wlen) + chunk = val_tokens[ws:end + 1].to(dtype=torch.int64, device=device) + x_batch[i, :wlen] = chunk[:-1] + y_batch[i, :wlen] = chunk[1:] + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + logits = compiled_logits(x_batch) + nll = F.cross_entropy( + logits.reshape(-1, logits.size(-1)).float(), + y_batch.reshape(-1), + reduction="none", + ).reshape(bsz, seq_len) + for i, ws in enumerate(batch_ws): + wlen = wlens[i] + s = 0 if ws == 0 else max(wlen - stride, 0) + scored_nll = nll[i, s:wlen].to(torch.float64) + loss_sum += scored_nll.sum() + token_count += float(wlen - s) + tgt = y_batch[i, s:wlen] + prev = x_batch[i, s:wlen] + tb = base_bytes_lut[tgt].to(torch.float64) + tb += (has_leading_space_lut[tgt] & ~is_boundary_token_lut[prev]).to(torch.float64) + byte_count += tb.sum() + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(token_count, op=dist.ReduceOp.SUM) + dist.all_reduce(byte_count, op=dist.ReduceOp.SUM) + val_loss = (loss_sum / token_count).item() + bits_per_token = val_loss / math.log(2.0) + tokens_per_byte = token_count.item() / byte_count.item() + base_model.train() + return val_loss, bits_per_token * tokens_per_byte + + +def ttt_adapt_adamw( + args, base_model: nn.Module, device: torch.device, + val_tokens: Tensor, rank: int = 0, world_size: int = 1, log0=print, +) -> None: + """AdamW TTT: fine-tune on val data BEFORE quantization (PR #1006 style).""" + seq_len = args.train_seq_len + total_seqs = (val_tokens.numel() - 1) // seq_len + batch_seqs = args.ttt_batch_seqs + if args.ttt_freeze_blocks > 0: + for i, block in enumerate(base_model.blocks): + if i < args.ttt_freeze_blocks: + for p in block.parameters(): + p.requires_grad_(False) + ttt_params = [p for p in base_model.parameters() if p.requires_grad] + log0(f"ttt_adamw:params trainable={sum(p.numel() for p in ttt_params)} " + f"frozen={sum(p.numel() for p in base_model.parameters() if not p.requires_grad)}") + optimizer = torch.optim.AdamW(ttt_params, lr=args.ttt_lr, weight_decay=0.0) + scheduler = None + if args.ttt_cosine_decay: + scheduler = torch.optim.lr_scheduler.CosineAnnealingLR( + optimizer, T_max=args.ttt_epochs, eta_min=args.ttt_lr * 0.1) + my_start = (total_seqs * rank) // world_size + my_end = (total_seqs * (rank + 1)) // world_size + base_model.train() + t0 = time.perf_counter() + for epoch in range(args.ttt_epochs): + epoch_loss_sum = torch.zeros((), device=device, dtype=torch.float64) + epoch_tokens = torch.zeros((), device=device, dtype=torch.float64) + for bs in range(my_start, my_end, batch_seqs): + be = min(bs + batch_seqs, my_end) + raw_start = bs * seq_len + raw_end = be * seq_len + 1 + if raw_end > val_tokens.numel(): + continue + local = val_tokens[raw_start:raw_end].to(device=device, dtype=torch.int64) + x = local[:-1].reshape(-1, seq_len) + y = local[1:].reshape(-1, seq_len) + optimizer.zero_grad(set_to_none=True) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + loss = base_model(x, y) + loss.backward() + if world_size > 1: + for p in ttt_params: + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG) + torch.nn.utils.clip_grad_norm_(ttt_params, args.ttt_grad_clip) + optimizer.step() + epoch_loss_sum += loss.detach().to(torch.float64) * float(y.numel()) + epoch_tokens += float(y.numel()) + if world_size > 1: + dist.all_reduce(epoch_loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(epoch_tokens, op=dist.ReduceOp.SUM) + epoch_avg_loss = epoch_loss_sum.item() / max(epoch_tokens.item(), 1) + if scheduler is not None: + scheduler.step() + log0(f"ttt_adamw:epoch {epoch+1}/{args.ttt_epochs} loss:{epoch_avg_loss:.4f} " + f"time:{time.perf_counter() - t0:.1f}s") + for p in base_model.parameters(): + p.requires_grad_(True) + base_model.eval() + log0(f"ttt_adamw:done elapsed={time.perf_counter() - t0:.1f}s") + + +def generate_autoregressive_calib(model, device, num_seqs=64, seq_len=2048, + vocab_size=1024, temperature=0.8, batch_size=8, seed=42): + """Generate sequences autoregressively from the model for GPTQ calibration. + No external data accessed — fully self-contained.""" + model.eval() + rng = torch.Generator(device=device) + rng.manual_seed(seed) + all_tokens = [] + with torch.inference_mode(), torch.autocast(device_type="cuda", dtype=torch.bfloat16): + for batch_start in range(0, num_seqs, batch_size): + bs = min(batch_size, num_seqs - batch_start) + tokens = torch.randint(0, vocab_size, (bs, 1), device=device, generator=rng) + for pos in range(seq_len - 1): + logits = model.forward_logits(tokens) + next_logit = logits[:, -1, :] + probs = torch.softmax(next_logit / temperature, dim=-1) + next_tok = torch.multinomial(probs, 1, generator=rng) + tokens = torch.cat([tokens, next_tok], dim=1) + for i in range(bs): + all_tokens.append(tokens[i:i+1]) + return all_tokens + + +def collect_hessians_from_tokens(hessian_model, token_seqs, device): + """Collect H = X^T X from pre-generated token sequences.""" + hessians = {} + hooks = [] + for name, module in hessian_model.named_modules(): + if isinstance(module, CastedLinear): + param_name = name + ".weight" + cols = module.weight.shape[1] + hessians[param_name] = torch.zeros(cols, cols, dtype=torch.float32, device='cpu') + h = module.register_forward_hook(_make_input_hook(hessians, param_name)) + hooks.append(h) + hessian_model.eval() + with torch.inference_mode(), torch.autocast(device_type="cuda", dtype=torch.bfloat16): + for seq in token_seqs: + x = seq[:, :-1].to(device) + y = seq[:, 1:].to(device) + hessian_model(x, y) + for h in hooks: + h.remove() + num_batches = len(token_seqs) + for name in hessians: + H = hessians[name] + H /= num_batches + damp = 0.01 * torch.diag(H).mean().clamp_min(1e-6) + H += damp * torch.eye(H.shape[0]) + hessians[name] = H + return hessians + + +# --- GPTQ-lite int6 quantization --- + +def _classify_param(name: str) -> str: + if "tok_emb" in name or "lm_head" in name: + return "embed" + if ".mlp." in name: + return "mlp" + if ".attn." in name or (".proj." in name and ".mlp." not in name): + return "attn" + return "other" +def sdclip_scale(t32: Tensor, clip_sigmas: float, clip_range: int) -> Tensor: + """SDClip: scale = clip_sigmas * std(row) / clip_range (per-row for 2D, global for 1D).""" + if t32.ndim == 2: + row_std = t32.std(dim=1).clamp_min(1e-10) + return (clip_sigmas * row_std / clip_range).clamp_min(1.0 / clip_range).to(torch.float16) + amax = t32.abs().max().item() + return torch.tensor(amax / clip_range if amax > 0 else 1.0, dtype=torch.float16) + +def quantize_int6_per_row(t: Tensor, clip_range: int = 31, clip_sigmas: float = 12.85) -> tuple[Tensor, Tensor]: + t32 = t.float() + s = sdclip_scale(t32, clip_sigmas, clip_range) + if t32.ndim == 2: + q = torch.clamp(torch.round(t32 / s.float()[:, None]), -clip_range, clip_range).to(torch.int8) + else: + q = torch.clamp(torch.round(t32 / s.float()), -clip_range, clip_range).to(torch.int8) + return q, s + +def quantize_int6_gptq(weight, hessian=None, clip_range=31, block_size=128, clip_sigmas=12.85): + """Full GPTQ: Hessian-aware int6 quantization with Cholesky error compensation. + Uses SDClip (k * std per row) for scale computation. + If hessian is None, falls back to SDClip-only quantization.""" + t32 = weight.float() + if t32.ndim != 2 or hessian is None: + return quantize_int6_per_row(t32, clip_range=clip_range, clip_sigmas=clip_sigmas) + rows, cols = t32.shape + H = hessian.float().clone() + dead = torch.diag(H) == 0 + H[dead, dead] = 1 + damp = 0.01 * torch.mean(torch.diag(H)) + H[torch.arange(cols), torch.arange(cols)] += damp + perm = torch.argsort(torch.diag(H), descending=True) + inv_perm = torch.argsort(perm) + W = t32[:, perm].clone() + W[:, dead[perm]] = 0 + H = H[perm][:, perm] + Hinv = torch.linalg.cholesky(H) + Hinv = torch.cholesky_inverse(Hinv) + Hinv = torch.linalg.cholesky(Hinv, upper=True) + # SDClip: single-pass scale from k * std(row) + s = sdclip_scale(t32, clip_sigmas, clip_range) + sf = s.float() + Q = torch.zeros_like(W, dtype=torch.int8) + W_work = W.clone() + for i1 in range(0, cols, block_size): + i2 = min(i1 + block_size, cols) + count = i2 - i1 + W1 = W_work[:, i1:i2].clone() + Q1 = torch.zeros(rows, count, dtype=torch.int8) + Err1 = torch.zeros(rows, count) + Hinv1 = Hinv[i1:i2, i1:i2] + for i in range(count): + w = W1[:, i] + d = Hinv1[i, i] + q = torch.clamp(torch.round(w / sf), -clip_range, clip_range).to(torch.int8) + Q1[:, i] = q + err = (w - q.float() * sf) / d + W1[:, i:] -= err.unsqueeze(1) * Hinv1[i, i:].unsqueeze(0) + Err1[:, i] = err + Q[:, i1:i2] = Q1 + if i2 < cols: + W_work[:, i2:] -= Err1 @ Hinv[i1:i2, i2:] + Q = Q[:, inv_perm] + return Q, s + +def _quantize_int6_percentile(t32, clip_range=31, clip_sigmas=12.85): + """Fallback: SDClip-based (kept for API compat, now uses SDClip not percentile).""" + return quantize_int6_per_row(t32, clip_range=clip_range, clip_sigmas=clip_sigmas) + +def _unbank_state_dict(sd: dict[str, Tensor], num_layers: int) -> dict[str, Tensor]: + """Convert 3D bank tensors into individual 2D tensors with standard names.""" + out: dict[str, Tensor] = {} + n = num_layers + for name, tensor in sd.items(): + if name == "qo_bank": + for i in range(n): + out[f"blocks.{i}.attn.c_q.weight"] = tensor[i] + out[f"blocks.{i}.attn.proj.weight"] = tensor[n + i] + elif name == "kv_bank": + for i in range(n): + out[f"blocks.{i}.attn.c_k.weight"] = tensor[i] + out[f"blocks.{i}.attn.c_v.weight"] = tensor[n + i] + elif name == "mlp_up_bank": + for i in range(n): + out[f"blocks.{i}.mlp.fc.weight"] = tensor[i] + elif name == "mlp_down_bank": + for i in range(n): + out[f"blocks.{i}.mlp.proj.weight"] = tensor[i] + else: + out[name] = tensor + return out + +def _rebank_state_dict(sd: dict[str, Tensor], num_layers: int, template_sd: dict[str, Tensor]) -> dict[str, Tensor]: + """Convert individual 2D tensors back into 3D bank tensors.""" + out: dict[str, Tensor] = {} + n = num_layers + # Reconstruct banks from individual weight keys + qo_slices = [None] * (2 * n) + kv_slices = [None] * (2 * n) + up_slices = [None] * n + down_slices = [None] * n + consumed = set() + for i in range(n): + qk = f"blocks.{i}.attn.c_q.weight" + if qk in sd: + qo_slices[i] = sd[qk] + consumed.add(qk) + ok = f"blocks.{i}.attn.proj.weight" + if ok in sd: + qo_slices[n + i] = sd[ok] + consumed.add(ok) + kk = f"blocks.{i}.attn.c_k.weight" + if kk in sd: + kv_slices[i] = sd[kk] + consumed.add(kk) + vk = f"blocks.{i}.attn.c_v.weight" + if vk in sd: + kv_slices[n + i] = sd[vk] + consumed.add(vk) + fk = f"blocks.{i}.mlp.fc.weight" + if fk in sd: + up_slices[i] = sd[fk] + consumed.add(fk) + dk = f"blocks.{i}.mlp.proj.weight" + if dk in sd: + down_slices[i] = sd[dk] + consumed.add(dk) + out["qo_bank"] = torch.stack(qo_slices).to(dtype=template_sd["qo_bank"].dtype) + out["kv_bank"] = torch.stack(kv_slices).to(dtype=template_sd["kv_bank"].dtype) + out["mlp_up_bank"] = torch.stack(up_slices).to(dtype=template_sd["mlp_up_bank"].dtype) + out["mlp_down_bank"] = torch.stack(down_slices).to(dtype=template_sd["mlp_down_bank"].dtype) + for name, tensor in sd.items(): + if name not in consumed: + out[name] = tensor + return out + +# --- Non-banked model for Hessian collection --- +# This mirrors the unbanked state dict keys: blocks.{i}.attn.c_q/c_k/c_v/proj, blocks.{i}.mlp.fc/proj + +class _HessianAttn(nn.Module): + """Non-banked attention with CastedLinear layers for Hessian hooks.""" + def __init__(self, dim, num_heads, num_kv_heads, rope_base, qk_gain_init): + super().__init__() + self.num_heads, self.num_kv_heads = num_heads, num_kv_heads + self.head_dim = dim // num_heads + kv_dim = num_kv_heads * self.head_dim + self.c_q = CastedLinear(dim, dim, bias=False) + self.c_k = CastedLinear(dim, kv_dim, bias=False) + self.c_v = CastedLinear(dim, kv_dim, bias=False) + self.proj = CastedLinear(dim, dim, bias=False) + self.q_gain = nn.Parameter(torch.full((num_heads,), qk_gain_init, dtype=torch.float32)) + self.rope_dims = 0 + self.rotary = Rotary(self.head_dim, base=rope_base, train_seq_len=1024) + self.use_xsa = False + def _xsa_efficient(self, y, v): + B, T, H, D = y.shape; Hkv = v.size(-2); group = H // Hkv + y_g = y.reshape(B, T, Hkv, group, D) + vn = F.normalize(v, dim=-1).unsqueeze(-2) + proj = (y_g * vn).sum(dim=-1, keepdim=True) * vn + return (y_g - proj).reshape(B, T, H, D) + def forward(self, x, v_embed=None): + bsz, seqlen, dim = x.shape + q = self.c_q(x).reshape(bsz, seqlen, self.num_heads, self.head_dim) + k = self.c_k(x).reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + v = self.c_v(x) + if v_embed is not None: + v = v + v_embed + v = v.reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + q = F.rms_norm(q, (q.size(-1),)) + k = F.rms_norm(k, (k.size(-1),)) + cos, sin = self.rotary(seqlen, x.device, q.dtype) + q = apply_rotary_emb(q, cos, sin, self.rope_dims) + k = apply_rotary_emb(k, cos, sin, self.rope_dims) + q = q * self.q_gain.to(dtype=q.dtype)[None, None, :, None] + if HAS_FA3: + y = flash_attn_3_func(q, k, v, causal=True) + else: + q2, k2, v2 = q.transpose(1, 2), k.transpose(1, 2), v.transpose(1, 2) + if k2.size(1) != q2.size(1): + reps = q2.size(1) // k2.size(1) + k2 = k2.repeat_interleave(reps, dim=1) + v2 = v2.repeat_interleave(reps, dim=1) + y = F.scaled_dot_product_attention(q2, k2, v2, is_causal=True).transpose(1, 2) + if self.use_xsa: + y = self._xsa_efficient(y, v) + return self.proj(y.reshape(bsz, seqlen, dim)) + +class _HessianMLP(nn.Module): + """Non-banked MLP with CastedLinear layers for Hessian hooks.""" + def __init__(self, dim, mlp_mult): + super().__init__() + self.fc = CastedLinear(dim, int(mlp_mult * dim), bias=False) + self.proj = CastedLinear(int(mlp_mult * dim), dim, bias=False) + def forward(self, x): + return self.proj(F.leaky_relu(self.fc(x), negative_slope=0.5).square()) + +class _HessianBlock(nn.Module): + def __init__(self, dim, num_heads, num_kv_heads, mlp_mult, rope_base, qk_gain_init, layer_idx=0, ln_scale=False): + super().__init__() + self.attn_norm = RMSNorm() + self.mlp_norm = RMSNorm() + self.attn = _HessianAttn(dim, num_heads, num_kv_heads, rope_base, qk_gain_init) + self.mlp = _HessianMLP(dim, mlp_mult) + self.attn_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.mlp_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.resid_mix = nn.Parameter(torch.stack((torch.ones(dim), torch.zeros(dim))).float()) + self.ln_scale_factor = 1.0 / math.sqrt(layer_idx + 1) if ln_scale else 1.0 + def forward(self, x, x0, v_embed=None): + mix = self.resid_mix.to(dtype=x.dtype) + x_in = mix[0][None, None, :] * x + mix[1][None, None, :] * x0 + attn_out = self.attn(self.attn_norm(x_in) * self.ln_scale_factor, v_embed=v_embed) + x_out = x_in + self.attn_scale.to(dtype=x_in.dtype)[None, None, :] * attn_out + x_out = x_out + self.mlp_scale.to(dtype=x_out.dtype)[None, None, :] * self.mlp(self.mlp_norm(x_out) * self.ln_scale_factor) + return x_out + +class _HessianGPT(nn.Module): + """Non-banked GPT model matching unbanked state dict keys for Hessian collection.""" + def __init__(self, vocab_size, num_layers, model_dim, num_heads, num_kv_heads, + mlp_mult, tie_embeddings, logit_softcap, rope_base, qk_gain_init, + bigram_vocab_size=0, bigram_dim=128, xsa_last_n=0, + rope_dims=0, ln_scale=False, + ve_enabled=False, ve_dim=128, ve_layers="9,10"): + super().__init__() + self.tie_embeddings = tie_embeddings + self.logit_softcap = logit_softcap + self.num_layers = num_layers + self.tok_emb = nn.Embedding(vocab_size, model_dim) + self.bigram = BigramHashEmbedding(bigram_vocab_size, bigram_dim, model_dim, trigram=bool(int(os.environ.get("TRIGRAM", "0")))) if bigram_vocab_size > 0 else None + self.smear = SmearGate(model_dim) + self.num_encoder_layers = num_layers // 2 + self.num_decoder_layers = num_layers - self.num_encoder_layers + self.num_skip_weights = min(self.num_encoder_layers, self.num_decoder_layers) + self.skip_weights = nn.Parameter(torch.ones(self.num_skip_weights, model_dim, dtype=torch.float32)) + self.blocks = nn.ModuleList([ + _HessianBlock(model_dim, num_heads, num_kv_heads, mlp_mult, rope_base, qk_gain_init, + layer_idx=i, ln_scale=ln_scale) + for i in range(num_layers) + ]) + if rope_dims > 0: + head_dim = model_dim // num_heads + for block in self.blocks: + block.attn.rope_dims = rope_dims + block.attn.rotary = Rotary(head_dim, base=rope_base, train_seq_len=1024, rope_dims=rope_dims) + if xsa_last_n > 0: + for i in range(max(0, num_layers - xsa_last_n), num_layers): + self.blocks[i].attn.use_xsa = True + kv_dim = num_kv_heads * (model_dim // num_heads) + self.ve_layer_indices = [int(x) for x in ve_layers.split(",") if x.strip()] if ve_enabled else [] + if self.ve_layer_indices: + self.ve_shared = ValueEmbedding(vocab_size, ve_dim, kv_dim) + self.ve_layer_scales = nn.ParameterList([nn.Parameter(torch.ones(1, dtype=torch.float32)) for _ in self.ve_layer_indices]) + else: + self.ve_shared = None + self.ve_layer_scales = nn.ParameterList() + self.final_norm = RMSNorm() + self.lm_head = None if tie_embeddings else CastedLinear(model_dim, vocab_size, bias=False) + def _get_ve(self, layer_idx, input_ids, ve_cache): + if self.ve_shared is None or layer_idx not in self.ve_layer_indices: + return None + if 've' not in ve_cache: + ve_cache['ve'] = self.ve_shared(input_ids) + ve_idx = self.ve_layer_indices.index(layer_idx) + return ve_cache['ve'] * self.ve_layer_scales[ve_idx].to(dtype=ve_cache['ve'].dtype) + def forward(self, input_ids, target_ids): + x = self.tok_emb(input_ids) + if self.bigram is not None: + x = x + self.bigram(input_ids) + x = F.rms_norm(x, (x.size(-1),)) + x = self.smear(x) + x0 = x + skips = [] + ve_cache = {} + for i in range(self.num_encoder_layers): + ve = self._get_ve(i, input_ids, ve_cache) + x = self.blocks[i](x, x0, v_embed=ve) + skips.append(x) + for i in range(self.num_decoder_layers): + bi = self.num_encoder_layers + i + if skips: + x = x + self.skip_weights[i].to(dtype=x.dtype)[None, None, :] * skips.pop() + ve = self._get_ve(bi, input_ids, ve_cache) + x = self.blocks[bi](x, x0, v_embed=ve) + x = self.final_norm(x) + x_flat = x.reshape(-1, x.size(-1)) + targets = target_ids.reshape(-1) + logits_proj = F.linear(x_flat, self.tok_emb.weight) if self.tie_embeddings else self.lm_head(x_flat) + logits = self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + return F.cross_entropy(logits.float(), targets, reduction="mean") + +def _make_input_hook(hessians, pname): + def hook_fn(module, input, output): + x = input[0].detach().float() + if x.ndim == 3: + x = x.reshape(-1, x.shape[-1]) + hessians[pname] += (x.T @ x).cpu() + return hook_fn + +def _make_output_hook(hessians, pname): + def hook_fn(module, input, output): + x = output.detach().float() + if x.ndim == 3: + x = x.reshape(-1, x.shape[-1]) + hessians[pname] += (x.T @ x).cpu() + return hook_fn + +def collect_hessians(hessian_model, train_loader, args, device, grad_accum_steps, num_batches=256): + """Run calibration batches through a non-banked model, collecting H = X^T X for each CastedLinear. + Also captures the pre-logit hidden states for tok_emb.weight GPTQ (embedding GPTQ).""" + hessians = {} + hooks = [] + for name, module in hessian_model.named_modules(): + if isinstance(module, CastedLinear): + param_name = name + ".weight" + cols = module.weight.shape[1] + hessians[param_name] = torch.zeros(cols, cols, dtype=torch.float32, device='cpu') + h = module.register_forward_hook(_make_input_hook(hessians, param_name)) + hooks.append(h) + # Embedding GPTQ: capture Hessian for tok_emb.weight via pre-logit hidden states. + # The tied embedding acts as a linear layer: logits = hidden @ tok_emb.weight.T + # so the activation Hessian is X^T X where X = final_norm output. + embed_hessian_key = "tok_emb.weight" + embed_dim = hessian_model.tok_emb.weight.shape[1] + hessians[embed_hessian_key] = torch.zeros(embed_dim, embed_dim, dtype=torch.float32, device='cpu') + embed_hook = hessian_model.final_norm.register_forward_hook( + _make_output_hook(hessians, embed_hessian_key) + ) + hooks.append(embed_hook) + hessian_model.eval() + with torch.inference_mode(), torch.autocast(device_type="cuda", dtype=torch.bfloat16): + for _ in range(num_batches): + x, y = train_loader.next_batch(args.train_batch_tokens, args.train_seq_len, grad_accum_steps) + hessian_model(x, y) + for h in hooks: + h.remove() + for name in hessians: + H = hessians[name] + H /= num_batches + damp = 0.01 * torch.diag(H).mean().clamp_min(1e-6) + H += damp * torch.eye(H.shape[0]) + hessians[name] = H + hessian_model.train() + return hessians + +def mixed_quantize_int6(state_dict: dict[str, Tensor], int6_cats: set[str], hessians: dict[str, Tensor] | None = None, + matrix_bits: int = 6, embed_bits: int = 8, + matrix_clip_sigmas: float = 12.85, embed_clip_sigmas: float = 20.0): + """Mixed-precision GPTQ quantization. + - mlp/attn weights: int6 with SDClip (k=matrix_clip_sigmas) and Hessian compensation + - tok_emb/lm_head: GPTQ with int8 and SDClip (k=embed_clip_sigmas) + - small tensors: passthrough float16 + """ + num_layers_total = max( + (int(k.split(".")[1]) for k in state_dict if k.startswith("blocks.")), + default=0, + ) + 1 + result: dict[str, Tensor] = {} + meta: dict[str, object] = {} + for name, tensor in state_dict.items(): + t = tensor.detach().cpu().contiguous() + cat = _classify_param(name) + if not t.is_floating_point() or t.numel() <= 65536: + result[name] = t.to(torch.float16) if t.is_floating_point() else t + meta[name] = "passthrough" + continue + if any(p in name for p in CONTROL_TENSOR_NAME_PATTERNS): + result[name] = t.float() + meta[name] = "passthrough_ctrl" + continue + if cat == "embed" and t.ndim >= 1: + # GPTQ on embeddings with int8 + SDClip + cr = 2 ** (embed_bits - 1) - 1 # 127 for int8 + H = hessians.get(name) if hessians else None + if H is not None: + q, s = quantize_int6_gptq(t, hessian=H, clip_range=cr, clip_sigmas=embed_clip_sigmas) + else: + q, s = quantize_int6_per_row(t, clip_range=cr, clip_sigmas=embed_clip_sigmas) + result[name + ".q"] = q + result[name + ".scale"] = s + meta[name] = {"type": f"int{embed_bits}"} + elif cat in int6_cats and t.ndim >= 1: + cr = 2 ** (matrix_bits - 1) - 1 # 31 for int6 + H = hessians.get(name) if hessians else None + if H is not None: + q, s = quantize_int6_gptq(t, hessian=H, clip_range=cr, clip_sigmas=matrix_clip_sigmas) + else: + q, s = quantize_int6_per_row(t, clip_range=cr, clip_sigmas=matrix_clip_sigmas) + result[name + ".q"] = q + result[name + ".scale"] = s + meta[name] = {"type": f"int{matrix_bits}"} + else: + q, s = quantize_float_tensor(t) + result[name + ".q"] = q + result[name + ".scale"] = s + meta[name] = {"type": "int8"} + return result, meta +def dequantize_mixed_int6(result: dict[str, Tensor], meta: dict[str, object], + template_sd: dict[str, Tensor]) -> dict[str, Tensor]: + out: dict[str, Tensor] = {} + for name, orig in template_sd.items(): + info = meta.get(name) + if info is None: + continue + orig_dtype = orig.dtype + if info in ("passthrough", "passthrough_ctrl", "passthrough_fp16"): + t = result[name] + if t.dtype == torch.float16 and orig_dtype in (torch.float32, torch.bfloat16): + t = t.to(orig_dtype) + out[name] = t + continue + q, s = result[name + ".q"], result[name + ".scale"] + if s.ndim > 0: + out[name] = (q.float() * s.float().view(q.shape[0], *([1] * (q.ndim - 1)))).to(orig_dtype) + else: + out[name] = (q.float() * float(s.item())).to(orig_dtype) + return out + +# --- Byte-shuffle + Brotli compression --- + +_BSHF_MAGIC = b'BSHF' + +def _byte_shuffle(data: bytes, stride: int = 2) -> bytes: + """Interleave bytes by stride before compression (improves Brotli ratio on quantized weights).""" + if stride <= 1 or len(data) < stride: + return data + src = np.frombuffer(data, dtype=np.uint8) + n = len(src) + out = np.empty(n, dtype=np.uint8) + dest_off = 0 + for pos in range(stride): + chunk = src[pos::stride] + out[dest_off:dest_off + len(chunk)] = chunk + dest_off += len(chunk) + return _BSHF_MAGIC + bytes([stride]) + out.tobytes() + +def _byte_unshuffle(data: bytes) -> bytes: + """Reverse byte shuffle.""" + if len(data) < 5 or data[:4] != _BSHF_MAGIC: + return data + stride = data[4] + if stride < 2: + return data[5:] + payload = np.frombuffer(data, dtype=np.uint8, offset=5) + n = len(payload) + out = np.empty(n, dtype=np.uint8) + src_off = 0 + for pos in range(stride): + chunk_len = n // stride + (1 if pos < n % stride else 0) + out[pos::stride][:chunk_len] = payload[src_off:src_off + chunk_len] + src_off += chunk_len + return out.tobytes() + +def _compress_brotli(data: bytes) -> bytes: + """Byte-shuffle then Brotli compress.""" + import brotli + shuffled = _byte_shuffle(data, stride=2) + return brotli.compress(shuffled, quality=11) + +def _decompress_brotli(data: bytes) -> bytes: + """Brotli decompress then byte-unshuffle.""" + import brotli + raw = brotli.decompress(data) + return _byte_unshuffle(raw) + +# --- Training --- + +def main() -> None: + code = Path(__file__).read_text(encoding="utf-8") + args = Hyperparameters() + # zeropower_via_newtonschulz5 runs eagerly with bmm -- do NOT compile + distributed = "RANK" in os.environ and "WORLD_SIZE" in os.environ + rank = int(os.environ.get("RANK", "0")) + world_size = int(os.environ.get("WORLD_SIZE", "1")) + local_rank = int(os.environ.get("LOCAL_RANK", "0")) + if world_size <= 0: + raise ValueError(f"WORLD_SIZE must be positive, got {world_size}") + if 8 % world_size != 0: + raise ValueError(f"WORLD_SIZE={world_size} must divide 8 so grad_accum_steps stays integral") + grad_accum_steps = 8 // world_size + grad_scale = 1.0 / grad_accum_steps + if not torch.cuda.is_available(): + raise RuntimeError("CUDA is required") + device = torch.device("cuda", local_rank) + torch.cuda.set_device(device) + if distributed: + dist.init_process_group(backend="nccl", device_id=device) + dist.barrier() + master_process = rank == 0 + torch.backends.cuda.matmul.allow_tf32 = True + torch.backends.cudnn.allow_tf32 = True + from torch.backends.cuda import enable_cudnn_sdp, enable_flash_sdp, enable_math_sdp, enable_mem_efficient_sdp + enable_cudnn_sdp(False) + enable_flash_sdp(True) + enable_mem_efficient_sdp(False) + enable_math_sdp(False) + logfile = None + if master_process: + os.makedirs("logs", exist_ok=True) + logfile = f"logs/{args.run_id}.txt" + print(logfile) + def log0(msg: str, console: bool = True) -> None: + if not master_process: + return + if console: + print(msg) + if logfile is not None: + with open(logfile, "a", encoding="utf-8") as f: + print(msg, file=f) + log0(code, console=False) + log0("=" * 100, console=False) + log0(f"Running Python {sys.version}", console=False) + log0(f"Running PyTorch {torch.__version__}", console=False) + log0( + subprocess.run(["nvidia-smi"], stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, check=False).stdout, + console=False, + ) + log0("=" * 100, console=False) + random.seed(args.seed) + np.random.seed(args.seed) + torch.manual_seed(args.seed) + torch.cuda.manual_seed_all(args.seed) + if not args.tokenizer_path.endswith(".model"): + raise ValueError(f"Script only setup for SentencePiece .model file: {args.tokenizer_path}") + sp = spm.SentencePieceProcessor(model_file=args.tokenizer_path) + if int(sp.vocab_size()) != args.vocab_size: + raise ValueError( + f"VOCAB_SIZE={args.vocab_size} does not match tokenizer vocab_size={int(sp.vocab_size())}" + ) + dataset_dir = Path(args.data_path).resolve() + actual_train_files = len(list(dataset_dir.glob("fineweb_train_*.bin"))) + effective_eval_seq_len = args.eval_seq_len if args.eval_seq_len > 0 else args.train_seq_len + val_seq_len = max(args.train_seq_len, effective_eval_seq_len) + val_tokens = load_validation_tokens(args.val_files, val_seq_len) + base_bytes_lut, has_leading_space_lut, is_boundary_token_lut = build_sentencepiece_luts( + sp, args.vocab_size, device + ) + log0(f"val_bpb:enabled tokenizer_kind=sentencepiece tokenizer_path={args.tokenizer_path}") + log0(f"train_loader:dataset:{dataset_dir.name} train_shards:{actual_train_files}") + log0(f"val_loader:shards pattern={args.val_files} tokens:{val_tokens.numel() - 1}") + CastedLinear._qat_enabled = args.qat_enabled + base_model = GPT( + vocab_size=args.vocab_size, + num_layers=args.num_layers, + model_dim=args.model_dim, + num_heads=args.num_heads, + num_kv_heads=args.num_kv_heads, + mlp_mult=args.mlp_mult, + tie_embeddings=args.tie_embeddings, + tied_embed_init_std=args.tied_embed_init_std, + logit_softcap=args.logit_softcap, + rope_base=args.rope_base, + qk_gain_init=args.qk_gain_init, + mtp_num_heads=args.mtp_num_heads, + mtp_loss_weight=args.mtp_loss_weight, + bigram_vocab_size=args.bigram_vocab_size, + bigram_dim=args.bigram_dim, + xsa_last_n=args.xsa_last_n, + rope_dims=args.rope_dims, + ln_scale=args.ln_scale, + dtg=args.dtg_enabled, + ve_enabled=args.ve_enabled, + ve_dim=args.ve_dim, + ve_layers=args.ve_layers, + gated_attention=args.gated_attention, + value_residual=args.value_residual, + skip_gates_enabled=args.skip_gates_enabled, + ).to(device).bfloat16() + # Banks stay FP32 (like CastedLinear weights), cast to BF16 in forward + base_model.qo_bank.data = base_model.qo_bank.data.float() + base_model.kv_bank.data = base_model.kv_bank.data.float() + base_model.mlp_up_bank.data = base_model.mlp_up_bank.data.float() + base_model.mlp_down_bank.data = base_model.mlp_down_bank.data.float() + for module in base_model.modules(): + if isinstance(module, CastedLinear): + module.float() + restore_low_dim_params_to_fp32(base_model) + # No DDP -- Parallel Muon handles bank grad communication via reduce-scatter, + # and non-bank grads are manually all-reduced before Adam steps. + compiled_model = torch.compile(base_model, dynamic=False, fullgraph=True) + model = compiled_model + + # Optimizer split: + # - 4 parameter banks -> Muon (batched Newton-Schulz) + # - token embedding -> Adam + # - scalars/control tensors -> Adam + # - bigram proj, mtp heads, VE proj -> Adam (small matrix params not worth banking) + matrix_params = [ + base_model.qo_bank, base_model.kv_bank, + base_model.mlp_up_bank, base_model.mlp_down_bank, + ] + block_named_params = list(base_model.blocks.named_parameters()) + scalar_params = [ + p + for name, p in block_named_params + if p.ndim < 2 or any(pattern in name for pattern in CONTROL_TENSOR_NAME_PATTERNS) + ] + if base_model.skip_weights.numel() > 0: + scalar_params.append(base_model.skip_weights) + if base_model.skip_gates is not None and base_model.skip_gates.numel() > 0: + scalar_params.append(base_model.skip_gates) + scalar_params.append(base_model.smear.gate) + if base_model.bigram is not None: + scalar_params.append(base_model.bigram.scale) + token_lr = args.tied_embed_lr if args.tie_embeddings else args.embed_lr + tok_params = [{"params": [base_model.tok_emb.weight], "lr": token_lr, "base_lr": token_lr}] + if base_model.bigram is not None: + tok_params.append({"params": [base_model.bigram.embed.weight], "lr": token_lr, "base_lr": token_lr}) + if base_model.bigram.proj is not None: + scalar_params.append(base_model.bigram.proj.weight) + if base_model.ve_shared is not None: + tok_params.append({"params": [base_model.ve_shared.embed.weight], "lr": token_lr, "base_lr": token_lr}) + if base_model.ve_shared.proj is not None: + scalar_params.append(base_model.ve_shared.proj.weight) + scalar_params.append(base_model.ve_shared.scale) + for s in base_model.ve_layer_scales: + scalar_params.append(s) + optimizer_tok = torch.optim.AdamW( + tok_params, + betas=(args.beta1, args.beta2), + eps=args.adam_eps, + weight_decay=args.embed_wd, + fused=True, + ) + optimizer_muon = Muon( + matrix_params, + lr=args.matrix_lr, + momentum=args.muon_momentum, + backend_steps=args.muon_backend_steps, + weight_decay=args.muon_wd, + ) + for group in optimizer_muon.param_groups: + group["base_lr"] = args.matrix_lr + optimizer_scalar = torch.optim.AdamW( + [{"params": scalar_params, "lr": args.scalar_lr, "base_lr": args.scalar_lr}], + betas=(args.beta1, args.beta2), + eps=args.adam_eps, + weight_decay=args.adam_wd, + fused=True, + ) + # Non-bank params that need manual all-reduce (replicated across GPUs) + replicated_params = list(optimizer_tok.param_groups[0]["params"]) + for pg in optimizer_tok.param_groups[1:]: + replicated_params.extend(pg["params"]) + replicated_params.extend(scalar_params) + + optimizer_head = None + if base_model.lm_head is not None: + optimizer_head = torch.optim.Adam( + [{"params": [base_model.lm_head.weight], "lr": args.head_lr, "base_lr": args.head_lr}], + betas=(args.beta1, args.beta2), + eps=args.adam_eps, + fused=True, + ) + replicated_params.append(base_model.lm_head.weight) + optimizers: list[torch.optim.Optimizer] = [optimizer_tok, optimizer_muon, optimizer_scalar] + if optimizer_head is not None: + optimizers.append(optimizer_head) + n_params = sum(p.numel() for p in base_model.parameters()) + mtp_params = sum(p.numel() for p in base_model.mtp_heads.parameters()) + log0(f"model_params:{n_params}") + log0(f"mtp_num_heads:{args.mtp_num_heads} mtp_loss_weight:{args.mtp_loss_weight} mtp_params:{mtp_params}") + xsa_layers = [i for i, b in enumerate(base_model.blocks) if b.attn.use_xsa] + log0(f"XSA:last_{args.xsa_last_n} active_layers:{xsa_layers}") + log0(f"world_size:{world_size} grad_accum_steps:{grad_accum_steps}") + log0("sdp_backends:cudnn=False flash=True mem_efficient=False math=False") + log0(f"attention_mode:gqa num_heads:{args.num_heads} num_kv_heads:{args.num_kv_heads}") + log0( + f"tie_embeddings:{args.tie_embeddings} embed_lr:{token_lr} " + f"head_lr:{args.head_lr if base_model.lm_head is not None else 0.0} " + f"matrix_lr:{args.matrix_lr} scalar_lr:{args.scalar_lr}" + ) + log0( + f"train_batch_tokens:{args.train_batch_tokens} train_seq_len:{args.train_seq_len} " + f"iterations:{args.iterations} warmup_steps:{args.warmup_steps} " + f"max_wallclock_seconds:{args.max_wallclock_seconds:.3f}" + ) + log0(f"seed:{args.seed}") + train_loader = DistributedTokenLoader(args.train_files, rank, world_size, device) + def zero_grad_all() -> None: + for opt in optimizers: + opt.zero_grad(set_to_none=True) + max_wallclock_ms = 1000.0 * args.max_wallclock_seconds if args.max_wallclock_seconds > 0 else None + def training_frac(step: int, elapsed_ms: float) -> float: + if max_wallclock_ms is None: + return step / max(args.iterations, 1) + return elapsed_ms / max(max_wallclock_ms, 1e-9) + def lr_mul(step: int, elapsed_ms: float) -> float: + # warmdown_frac > 0 takes priority; fallback to legacy warmdown_iters + if args.warmdown_frac > 0: + frac = training_frac(step, elapsed_ms) + if frac >= 1.0 - args.warmdown_frac: + return max((1.0 - frac) / args.warmdown_frac, 0.0) + return 1.0 + if args.warmdown_iters <= 0: + return 1.0 + if max_wallclock_ms is None: + warmdown_start = max(args.iterations - args.warmdown_iters, 0) + return max((args.iterations - step) / max(args.warmdown_iters, 1), 0.0) if warmdown_start <= step < args.iterations else 1.0 + step_ms = elapsed_ms / max(step, 1) + warmdown_ms = args.warmdown_iters * step_ms + remaining_ms = max(max_wallclock_ms - elapsed_ms, 0.0) + return remaining_ms / max(warmdown_ms, 1e-9) if remaining_ms <= warmdown_ms else 1.0 + if args.warmup_steps > 0: + initial_model_state = {name: tensor.detach().cpu().clone() for name, tensor in base_model.state_dict().items()} + initial_optimizer_states = [copy.deepcopy(opt.state_dict()) for opt in optimizers] + model.train() + for warmup_step in range(args.warmup_steps): + zero_grad_all() + for micro_step in range(grad_accum_steps): + x, y = train_loader.next_batch(args.train_batch_tokens, args.train_seq_len, grad_accum_steps) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + warmup_loss = model(x, y) + (warmup_loss * grad_scale).backward() + # All-reduce all grads for warmup (simple, not optimized) + if distributed: + for p in base_model.parameters(): + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG) + for opt in optimizers: + opt.step() + zero_grad_all() + if args.warmup_steps <= 20 or (warmup_step + 1) % 10 == 0 or warmup_step + 1 == args.warmup_steps: + log0(f"warmup_step:{warmup_step + 1}/{args.warmup_steps}") + base_model.load_state_dict(initial_model_state, strict=True) + for opt, state in zip(optimizers, initial_optimizer_states, strict=True): + opt.load_state_dict(state) + zero_grad_all() + train_loader = DistributedTokenLoader(args.train_files, rank, world_size, device) + swa_state: dict[str, Tensor] | None = None + swa_count = 0 + from collections import deque + lawa_queue: deque[dict[str, Tensor]] = deque(maxlen=args.lawa_k) + ema_state = {name: t.detach().float().clone() for name, t in base_model.state_dict().items()} + ema_decay = 0.997 + training_time_ms = 0.0 + stop_after_step: int | None = None + torch.cuda.synchronize() + t0 = time.perf_counter() + step = 0 + while True: + last_step = step == args.iterations or (stop_after_step is not None and step >= stop_after_step) + should_validate = last_step or (args.val_loss_every > 0 and step % args.val_loss_every == 0) + if should_validate: + torch.cuda.synchronize() + training_time_ms += 1000.0 * (time.perf_counter() - t0) + val_loss, val_bpb = eval_val( + args, + model, + rank, + world_size, + device, + grad_accum_steps, + val_tokens, + base_bytes_lut, + has_leading_space_lut, + is_boundary_token_lut, + ) + log0( + f"step:{step}/{args.iterations} val_loss:{val_loss:.4f} val_bpb:{val_bpb:.4f} " + f"train_time:{training_time_ms:.0f}ms step_avg:{training_time_ms / max(step, 1):.2f}ms" + ) + torch.cuda.synchronize() + t0 = time.perf_counter() + if last_step: + if stop_after_step is not None and step < args.iterations: + log0( + f"stopping_early: wallclock_cap train_time:{training_time_ms:.0f}ms " + f"step:{step}/{args.iterations}" + ) + break + elapsed_ms = training_time_ms + 1000.0 * (time.perf_counter() - t0) + scale = lr_mul(step, elapsed_ms) + if args.late_qat_threshold > 0 and scale < args.late_qat_threshold and not CastedLinear._qat_enabled: + CastedLinear._qat_enabled = True + log0(f"late_qat:enabled step:{step} scale:{scale:.4f}") + zero_grad_all() + train_loss = torch.zeros((), device=device) + for micro_step in range(grad_accum_steps): + x, y = train_loader.next_batch(args.train_batch_tokens, args.train_seq_len, grad_accum_steps) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + loss = model(x, y) + train_loss += loss.detach() + (loss * grad_scale).backward() + train_loss /= grad_accum_steps + frac = min(step / args.muon_momentum_warmup_steps, 1.0) if args.muon_momentum_warmup_steps > 0 else 1.0 + muon_momentum = (1 - frac) * args.muon_momentum_warmup_start + frac * args.muon_momentum + for group in optimizer_muon.param_groups: + group["momentum"] = muon_momentum + for opt in optimizers: + for group in opt.param_groups: + group["lr"] = group["base_lr"] * scale + if args.grad_clip_norm > 0: + torch.nn.utils.clip_grad_norm_(base_model.parameters(), args.grad_clip_norm) + # === 3-phase overlapped optimizer step === + # Phase 1: Launch async reduce-scatter for banks (biggest first) + optimizer_muon.launch_reduce_scatters() + # Phase 2: All-reduce non-bank grads + step Adam (while bank RS is in-flight) + if distributed: + for p in replicated_params: + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG) + optimizer_tok.step() + optimizer_scalar.step() + if optimizer_head is not None: + optimizer_head.step() + # Phase 3: Wait for RS, local NS5, all-gather (banks processed last) + optimizer_muon.step() + zero_grad_all() + # EMA update + with torch.no_grad(): + for name, t in base_model.state_dict().items(): + ema_state[name].mul_(ema_decay).add_(t.detach().float(), alpha=1.0 - ema_decay) + step += 1 + approx_training_time_ms = training_time_ms + 1000.0 * (time.perf_counter() - t0) + if args.swa_enabled and scale < 0.2 and step % args.swa_every == 0: + if swa_state is None: + swa_state = {name: t.detach().cpu().clone() for name, t in base_model.state_dict().items()} + swa_count = 1 + log0(f"swa:start step:{step}") + else: + for name, t in base_model.state_dict().items(): + swa_state[name] += t.detach().cpu() + swa_count += 1 + if args.lawa_enabled and step % args.lawa_freq == 0: + lawa_queue.append({name: t.detach().cpu().clone() for name, t in base_model.state_dict().items()}) + should_log_train = ( + args.train_log_every > 0 + and (step <= 10 or step % args.train_log_every == 0 or stop_after_step is not None) + ) + if should_log_train: + log0( + f"step:{step}/{args.iterations} train_loss:{train_loss.item():.4f} " + f"train_time:{approx_training_time_ms:.0f}ms step_avg:{approx_training_time_ms / step:.2f}ms" + ) + reached_cap = max_wallclock_ms is not None and approx_training_time_ms >= max_wallclock_ms + if distributed and max_wallclock_ms is not None: + reached_cap_tensor = torch.tensor(int(reached_cap), device=device) + dist.all_reduce(reached_cap_tensor, op=dist.ReduceOp.MAX) + reached_cap = bool(reached_cap_tensor.item()) + if stop_after_step is None and reached_cap: + stop_after_step = step + log0( + f"peak memory allocated: {torch.cuda.max_memory_allocated() // 1024 // 1024} MiB " + f"reserved: {torch.cuda.max_memory_reserved() // 1024 // 1024} MiB" + ) + # Apply weight averaging + if args.lawa_enabled and len(lawa_queue) > 1: + log0(f"lawa:applying LAWA averaging k={len(lawa_queue)}") + current_state = base_model.state_dict() + avg_state = {name: torch.zeros(t.shape, dtype=torch.float32, device='cpu') for name, t in current_state.items()} + for snap in lawa_queue: + for name in avg_state: + avg_state[name] += snap[name].float() + for name in avg_state: + avg_state[name] /= len(lawa_queue) + avg_state[name] = avg_state[name].to(dtype=current_state[name].dtype) + base_model.load_state_dict(avg_state, strict=True) + else: + log0("ema:applying EMA weights") + current_state = base_model.state_dict() + avg_state = {name: t.to(dtype=current_state[name].dtype) for name, t in ema_state.items()} + base_model.load_state_dict(avg_state, strict=True) + torch.cuda.synchronize() + t_diag = time.perf_counter() + diag_val_loss, diag_val_bpb = eval_val( + args, compiled_model, rank, world_size, device, grad_accum_steps, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + ) + torch.cuda.synchronize() + log0( + f"DIAGNOSTIC post_ema val_loss:{diag_val_loss:.4f} val_bpb:{diag_val_bpb:.4f} " + f"eval_time:{1000.0 * (time.perf_counter() - t_diag):.0f}ms" + ) + # AdamW TTT: fine-tune EMA model on val data BEFORE quantization + if args.ttt_enabled: + if distributed: + dist.barrier() + log0(f"ttt:start lr={args.ttt_lr} epochs={args.ttt_epochs} " + f"freeze_blocks={args.ttt_freeze_blocks} cosine_decay={args.ttt_cosine_decay}") + t_ttt = time.perf_counter() + ttt_adapt_adamw( + args, base_model, device, val_tokens, + rank=rank, world_size=world_size, log0=log0, + ) + torch.cuda.synchronize() + log0(f"ttt:elapsed={time.perf_counter() - t_ttt:.1f}s") + t_ttt_diag = time.perf_counter() + ttt_diag_loss, ttt_diag_bpb = eval_val( + args, base_model, rank, world_size, device, grad_accum_steps, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + ) + torch.cuda.synchronize() + log0(f"DIAGNOSTIC post_ttt val_loss:{ttt_diag_loss:.4f} val_bpb:{ttt_diag_bpb:.4f} " + f"eval_time:{1000.0 * (time.perf_counter() - t_ttt_diag):.0f}ms") + if distributed: + dist.barrier() + full_state_dict = base_model.state_dict() + export_sd = {k: v for k, v in full_state_dict.items() if "mtp_heads" not in k} + excluded_mtp = sum(int(t.numel()) for k, t in full_state_dict.items() if "mtp_heads" in k) + if excluded_mtp > 0: + log0(f"export_excluding_mtp_params:{excluded_mtp}") + if master_process: + torch.save(export_sd, "final_model.pt") + model_bytes = os.path.getsize("final_model.pt") + code_bytes = len(code.encode("utf-8")) + log0(f"Serialized model: {model_bytes} bytes") + log0(f"Code size: {code_bytes} bytes") + # Unbank 3D tensors into individual 2D tensors for quantization + sd_cpu = {k: v.detach().cpu() for k, v in export_sd.items()} + unbanked_sd = _unbank_state_dict(sd_cpu, args.num_layers) + # Full GPTQ: collect Hessians via a temporary non-banked model + log0(f"gptq:building non-banked model for Hessian collection...") + hessian_model = _HessianGPT( + vocab_size=args.vocab_size, num_layers=args.num_layers, model_dim=args.model_dim, + num_heads=args.num_heads, num_kv_heads=args.num_kv_heads, mlp_mult=args.mlp_mult, + tie_embeddings=args.tie_embeddings, logit_softcap=args.logit_softcap, + rope_base=args.rope_base, qk_gain_init=args.qk_gain_init, + bigram_vocab_size=args.bigram_vocab_size, bigram_dim=args.bigram_dim, + xsa_last_n=args.xsa_last_n, rope_dims=args.rope_dims, ln_scale=args.ln_scale, + ve_enabled=args.ve_enabled, ve_dim=args.ve_dim, ve_layers=args.ve_layers, + ).to(device).bfloat16() + for m in hessian_model.modules(): + if isinstance(m, CastedLinear): + m.float() + restore_low_dim_params_to_fp32(hessian_model) + # Load unbanked weights into the non-banked model + hessian_model.load_state_dict( + {k: v.to(device) for k, v in unbanked_sd.items() if k in hessian_model.state_dict()}, + strict=False, + ) + # Training-data calibration (proven -0.0007 BPP vs AR self-gen) + log0(f"gptq:collecting hessians from training data ({args.gptq_calib_batches} batches)...") + hessians = collect_hessians(hessian_model, train_loader, args, device, grad_accum_steps, num_batches=args.gptq_calib_batches) + log0(f"gptq:collected hessians for {len(hessians)} layers (training data)") + del hessian_model + torch.cuda.empty_cache() + quant_result, quant_meta = mixed_quantize_int6( + unbanked_sd, {"mlp", "attn"}, hessians=hessians, + matrix_bits=args.matrix_bits, embed_bits=args.embed_bits, + matrix_clip_sigmas=args.matrix_clip_sigmas, embed_clip_sigmas=args.embed_clip_sigmas, + ) + # NOVEL: Selective ±1 pruning by reconstruction error + # Sort ±1 quantized values by their reconstruction error (scale²), + # prune least-impactful first until artifact fits target size. + target_mb = float(os.environ.get("TARGET_MB", "15.9")) + code_bytes_est = len(code.encode("utf-8")) + ones_info = [] # (tensor_key, flat_idx, error) + for name, info in quant_meta.items(): + if not (isinstance(info, dict) and info.get("type") == "int6"): continue + qk, sk = name + ".q", name + ".scale" + if qk not in quant_result or sk not in quant_result: continue + q, s = quant_result[qk], quant_result[sk] + if s.ndim > 0: + ones_mask = (q.abs() == 1) + if ones_mask.any(): + row_idx = torch.arange(q.shape[0]).unsqueeze(1).expand_as(q)[ones_mask] + flat_idx = torch.arange(q.numel()).reshape(q.shape)[ones_mask] + errors = s.float()[row_idx].pow(2) + for fi, err in zip(flat_idx.tolist(), errors.tolist()): + ones_info.append((qk, fi, err)) + if ones_info: + ones_info.sort(key=lambda x: x[2]) + def _try_prune(n): + tmp = {k: v.clone() for k, v in quant_result.items()} + for i in range(min(n, len(ones_info))): + tmp[ones_info[i][0]].view(-1)[ones_info[i][1]] = 0 + buf = io.BytesIO(); torch.save({"w": tmp, "m": quant_meta}, buf) + return len(_compress_brotli(buf.getvalue())) + code_bytes_est, tmp + no_sz, _ = _try_prune(0) + target_bytes = int(target_mb * 1024 * 1024) + log0(f"selective_prune: {len(ones_info)} ±1 candidates, unpruned={no_sz/(1024*1024):.2f}MB target={target_mb}MB") + if no_sz <= target_bytes: + log0("selective_prune: already fits, no pruning needed") + else: + full_sz, _ = _try_prune(len(ones_info)) + log0(f"selective_prune: full ±1 prune={full_sz/(1024*1024):.2f}MB") + if full_sz > target_bytes: + log0("selective_prune: even full prune not enough, applying all") + _, quant_result = _try_prune(len(ones_info)) + else: + lo, hi = 0, len(ones_info) + while lo < hi: + mid = (lo + hi) // 2 + sz, _ = _try_prune(mid) + if sz <= target_bytes: hi = mid + else: lo = mid + 1 + log0(f"selective_prune: pruning {lo}/{len(ones_info)} ±1 values ({100*lo/len(ones_info):.1f}%) to fit {target_mb}MB") + _, quant_result = _try_prune(lo) + quant_buf = io.BytesIO() + torch.save({"w": quant_result, "m": quant_meta}, quant_buf) + quant_raw = quant_buf.getvalue() + quant_blob = _compress_brotli(quant_raw) + if master_process: + with open("final_model.int6.ptz", "wb") as f: + f.write(quant_blob) + quant_file_bytes = len(quant_blob) + code_bytes = len(code.encode("utf-8")) + log0(f"Serialized model int6+brotli+byteshuffle: {quant_file_bytes} bytes") + log0(f"Total submission size int6+brotli+byteshuffle: {quant_file_bytes + code_bytes} bytes") + if distributed: + dist.barrier() + with open("final_model.int6.ptz", "rb") as f: + quant_blob_disk = f.read() + quant_state = torch.load( + io.BytesIO(_decompress_brotli(quant_blob_disk)), + map_location="cpu", + weights_only=False, + ) + deq_unbanked = dequantize_mixed_int6(quant_state["w"], quant_state["m"], unbanked_sd) + # Re-bank the dequantized tensors + deq_state = _rebank_state_dict(deq_unbanked, args.num_layers, sd_cpu) + eval_model = GPT( + vocab_size=args.vocab_size, num_layers=args.num_layers, model_dim=args.model_dim, + num_heads=args.num_heads, num_kv_heads=args.num_kv_heads, mlp_mult=args.mlp_mult, + tie_embeddings=args.tie_embeddings, tied_embed_init_std=args.tied_embed_init_std, + logit_softcap=args.logit_softcap, rope_base=args.rope_base, qk_gain_init=args.qk_gain_init, + mtp_num_heads=0, mtp_loss_weight=0.0, + bigram_vocab_size=args.bigram_vocab_size, bigram_dim=args.bigram_dim, + xsa_last_n=args.xsa_last_n, + rope_dims=args.rope_dims, ln_scale=args.ln_scale, dtg=args.dtg_enabled, + ve_enabled=args.ve_enabled, ve_dim=args.ve_dim, ve_layers=args.ve_layers, + gated_attention=args.gated_attention, value_residual=args.value_residual, + skip_gates_enabled=args.skip_gates_enabled, + ).to(device).bfloat16() + eval_model.qo_bank.data = eval_model.qo_bank.data.float() + eval_model.kv_bank.data = eval_model.kv_bank.data.float() + eval_model.mlp_up_bank.data = eval_model.mlp_up_bank.data.float() + eval_model.mlp_down_bank.data = eval_model.mlp_down_bank.data.float() + for m in eval_model.modules(): + if isinstance(m, CastedLinear): + m.float() + restore_low_dim_params_to_fp32(eval_model) + eval_model.load_state_dict(deq_state, strict=True) + compiled_eval = torch.compile(eval_model, dynamic=False, fullgraph=True) + torch.cuda.synchronize() + t_qeval = time.perf_counter() + q_val_loss, q_val_bpb = eval_val( + args, compiled_eval, rank, world_size, device, grad_accum_steps, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + eval_seq_len=effective_eval_seq_len, + ) + torch.cuda.synchronize() + log0( + f"final_int6_roundtrip val_loss:{q_val_loss:.4f} val_bpb:{q_val_bpb:.4f} " + f"eval_time:{1000.0 * (time.perf_counter() - t_qeval):.0f}ms" + ) + log0(f"final_int6_roundtrip_exact val_loss:{q_val_loss:.8f} val_bpb:{q_val_bpb:.8f}") + sw_seq_len = effective_eval_seq_len + if args.eval_stride > 0 and args.eval_stride < sw_seq_len: + torch.cuda.synchronize() + t_slide = time.perf_counter() + sw_val_loss, sw_val_bpb = eval_val_sliding( + args, eval_model, rank, world_size, device, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + stride=args.eval_stride, + eval_seq_len=sw_seq_len, + ) + torch.cuda.synchronize() + log0( + f"final_int6_sliding_window val_loss:{sw_val_loss:.4f} val_bpb:{sw_val_bpb:.4f} " + f"stride:{args.eval_stride} eval_time:{1000.0 * (time.perf_counter() - t_slide):.0f}ms" + ) + log0(f"final_int6_sliding_window_exact val_loss:{sw_val_loss:.8f} val_bpb:{sw_val_bpb:.8f}") + log0(f"final_int8_zlib_roundtrip_exact val_loss:{sw_val_loss:.8f} val_bpb:{sw_val_bpb:.8f}") + if args.eval_stride != 64 and 64 < sw_seq_len: + torch.cuda.synchronize() + t_slide64 = time.perf_counter() + sw64_val_loss, sw64_val_bpb = eval_val_sliding( + args, eval_model, rank, world_size, device, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + stride=64, + eval_seq_len=sw_seq_len, + ) + torch.cuda.synchronize() + log0( + f"final_int6_sliding_window_s64 val_loss:{sw64_val_loss:.4f} val_bpb:{sw64_val_bpb:.4f} " + f"stride:64 eval_time:{1000.0 * (time.perf_counter() - t_slide64):.0f}ms" + ) + log0(f"final_int6_sliding_window_s64_exact val_loss:{sw64_val_loss:.8f} val_bpb:{sw64_val_bpb:.8f}") + log0(f"final_int8_zlib_roundtrip_exact val_loss:{sw64_val_loss:.8f} val_bpb:{sw64_val_bpb:.8f}") + # --- Causal SLOT: per-batch delta optimization using ONLY context (already-scored) positions --- + if args.slot_enabled and args.eval_stride > 0 and args.eval_stride < sw_seq_len: + try: + slot_stride = args.eval_stride + seq_s = sw_seq_len + total_tok = val_tokens.numel() - 1 + ws_list = [ws for ws in range(0, total_tok, slot_stride) if min(ws + seq_s, total_tok) - ws >= 1] + my_s = (len(ws_list) * rank) // world_size + my_e = (len(ws_list) * (rank + 1)) // world_size + my_ws = ws_list[my_s:my_e] + num_batches = (len(my_ws) + 31) // 32 + sl_loss = torch.zeros((), device=device, dtype=torch.float64) + sl_tc = torch.zeros((), device=device, dtype=torch.float64) + sl_bc = torch.zeros((), device=device, dtype=torch.float64) + torch.cuda.synchronize() + t_slot = time.perf_counter() + eval_model.eval() + log0(f"causal_slot:start lr={args.slot_lr} steps={args.slot_steps} stride={slot_stride} " + f"windows={len(my_ws)} batches={num_batches}") + for batch_idx, bi in enumerate(range(0, len(my_ws), 32)): + bws = my_ws[bi:bi+32]; bsz = len(bws) + xb = torch.zeros(bsz, seq_s, dtype=torch.int64, device=device) + yb = torch.zeros(bsz, seq_s, dtype=torch.int64, device=device) + wls = [] + for i, ws in enumerate(bws): + end = min(ws + seq_s, total_tok); wl = end - ws; wls.append(wl) + ct = val_tokens[ws:end+1].to(dtype=torch.int64, device=device) + xb[i,:wl] = ct[:-1]; yb[i,:wl] = ct[1:] + # Frozen forward: get hidden states + with torch.no_grad(), torch.autocast(device_type="cuda", dtype=torch.bfloat16): + H = eval_model.forward_hidden(xb) + H = H.detach().float() + # Build context mask (already-scored positions) and score mask (new positions) + context_mask = torch.zeros(bsz, seq_s, dtype=torch.bool, device=device) + has_context = False + for i, ws in enumerate(bws): + wl = wls[i]; s = 0 if ws == 0 else max(wl - slot_stride, 0) + if s > 0: + context_mask[i, :s] = True + has_context = True + # Optimize delta on context positions only (CAUSAL: no future info) + delta = torch.zeros(1, 1, H.shape[-1], device=device, dtype=H.dtype, requires_grad=True) + if has_context: + sopt = torch.optim.AdamW([delta], lr=args.slot_lr, betas=(0.3, 0.9), weight_decay=1e-8, eps=1e-5) + for _ in range(args.slot_steps): + sopt.zero_grad() + lg = eval_model.compute_logits((H + delta).to(torch.bfloat16)).float() + nll_all = F.cross_entropy(lg.reshape(-1, lg.size(-1)), yb.reshape(-1), reduction="none").reshape(bsz, seq_s) + ctx_nll = nll_all[context_mask] + if ctx_nll.numel() > 0: + loss_c = ctx_nll.mean() + loss_c.backward() + sopt.step() + # Score new positions with adapted delta + with torch.no_grad(): + lg = eval_model.compute_logits((H + delta.detach()).to(torch.bfloat16)).float() + nll = F.cross_entropy(lg.reshape(-1, lg.size(-1)), yb.reshape(-1), reduction="none").reshape(bsz, seq_s) + for i, ws in enumerate(bws): + wl = wls[i]; s = 0 if ws == 0 else max(wl - slot_stride, 0) + sl_loss += nll[i, s:wl].to(torch.float64).sum(); sl_tc += float(wl - s) + tgt, prev = yb[i, s:wl], xb[i, s:wl] + tb = base_bytes_lut[tgt].to(torch.float64) + tb += (has_leading_space_lut[tgt] & ~is_boundary_token_lut[prev]).to(torch.float64) + sl_bc += tb.sum() + if batch_idx % 500 == 0 or batch_idx == num_batches - 1: + log0(f" causal_slot:batch {batch_idx+1}/{num_batches} " + f"time:{time.perf_counter()-t_slot:.1f}s"); sys.stdout.flush() + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(sl_loss, op=dist.ReduceOp.SUM) + dist.all_reduce(sl_tc, op=dist.ReduceOp.SUM) + dist.all_reduce(sl_bc, op=dist.ReduceOp.SUM) + sv_loss = (sl_loss / sl_tc).item() + sv_bpb = sv_loss / math.log(2.0) * (sl_tc.item() / sl_bc.item()) + torch.cuda.synchronize() + log0(f"final_causal_slot val_loss:{sv_loss:.4f} val_bpb:{sv_bpb:.4f} time:{1000*(time.perf_counter()-t_slot):.0f}ms") + log0(f"final_causal_slot_exact val_loss:{sv_loss:.8f} val_bpb:{sv_bpb:.8f}") + log0(f"final_int8_zlib_roundtrip_exact val_loss:{sv_loss:.8f} val_bpb:{sv_bpb:.8f}") + except Exception as e: + import traceback; log0(f"causal_slot:ERROR {e}"); traceback.print_exc(); sys.stdout.flush() + if distributed: + dist.destroy_process_group() + sys.exit(0) +if __name__ == "__main__": + main() + +==================================================================================================== +Running Python 3.12.3 (main, Nov 6 2025, 13:44:16) [GCC 13.3.0] +Running PyTorch 2.9.1+cu128 +Mon Apr 6 05:39:52 2026 ++-----------------------------------------------------------------------------------------+ +| NVIDIA-SMI 580.126.09 Driver Version: 580.126.09 CUDA Version: 13.0 | ++-----------------------------------------+------------------------+----------------------+ +| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC | +| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. | +| | | MIG M. | +|=========================================+========================+======================| +| 0 NVIDIA H100 80GB HBM3 On | 00000000:19:00.0 Off | 0 | +| N/A 45C P0 124W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 1 NVIDIA H100 80GB HBM3 On | 00000000:3B:00.0 Off | 0 | +| N/A 36C P0 120W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 2 NVIDIA H100 80GB HBM3 On | 00000000:4C:00.0 Off | 0 | +| N/A 34C P0 119W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 3 NVIDIA H100 80GB HBM3 On | 00000000:5D:00.0 Off | 0 | +| N/A 45C P0 125W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 4 NVIDIA H100 80GB HBM3 On | 00000000:9B:00.0 Off | 0 | +| N/A 46C P0 127W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 5 NVIDIA H100 80GB HBM3 On | 00000000:BB:00.0 Off | 0 | +| N/A 37C P0 120W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 6 NVIDIA H100 80GB HBM3 On | 00000000:CB:00.0 Off | 0 | +| N/A 44C P0 125W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 7 NVIDIA H100 80GB HBM3 On | 00000000:DB:00.0 Off | 0 | +| N/A 34C P0 112W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ + ++-----------------------------------------------------------------------------------------+ +| Processes: | +| GPU GI CI PID Type Process name GPU Memory | +| ID ID Usage | +|=========================================================================================| +| No running processes found | ++-----------------------------------------------------------------------------------------+ + +==================================================================================================== +val_bpb:enabled tokenizer_kind=sentencepiece tokenizer_path=./data/tokenizers/fineweb_8192_bpe.model +train_loader:dataset:fineweb10B_sp8192 train_shards:128 +val_loader:shards pattern=./data/datasets/fineweb10B_sp8192/fineweb_val_*.bin tokens:40540160 +model_params:35941976 +mtp_num_heads:0 mtp_loss_weight:0.2 mtp_params:0 +XSA:last_11 active_layers:[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10] +world_size:8 grad_accum_steps:1 +sdp_backends:cudnn=False flash=True mem_efficient=False math=False +attention_mode:gqa num_heads:8 num_kv_heads:4 +tie_embeddings:True embed_lr:0.035 head_lr:0.0 matrix_lr:0.025 scalar_lr:0.025 +train_batch_tokens:786432 train_seq_len:2048 iterations:20000 warmup_steps:20 max_wallclock_seconds:600.000 +seed:2025 +warmup_step:1/20 +warmup_step:2/20 +warmup_step:3/20 +warmup_step:4/20 +warmup_step:5/20 +warmup_step:6/20 +warmup_step:7/20 +warmup_step:8/20 +warmup_step:9/20 +warmup_step:10/20 +warmup_step:11/20 +warmup_step:12/20 +warmup_step:13/20 +warmup_step:14/20 +warmup_step:15/20 +warmup_step:16/20 +warmup_step:17/20 +warmup_step:18/20 +warmup_step:19/20 +warmup_step:20/20 +step:0/20000 val_loss:9.0006 val_bpb:3.4844 train_time:0ms step_avg:0.02ms +step:1/20000 train_loss:9.0029 train_time:143ms step_avg:143.39ms +step:2/20000 train_loss:9.7759 train_time:172ms step_avg:85.80ms +step:3/20000 train_loss:9.9749 train_time:268ms step_avg:89.33ms +step:4/20000 train_loss:9.4502 train_time:365ms step_avg:91.18ms +step:5/20000 train_loss:8.8628 train_time:462ms step_avg:92.49ms +step:6/20000 train_loss:8.6170 train_time:560ms step_avg:93.35ms +step:7/20000 train_loss:8.3200 train_time:659ms step_avg:94.20ms +step:8/20000 train_loss:7.8870 train_time:760ms step_avg:95.06ms +step:9/20000 train_loss:7.8087 train_time:860ms step_avg:95.57ms +step:10/20000 train_loss:7.6006 train_time:959ms step_avg:95.90ms +step:500/20000 train_loss:3.3400 train_time:49920ms step_avg:99.84ms +step:1000/20000 train_loss:3.0782 train_time:100061ms step_avg:100.06ms +step:1500/20000 train_loss:3.2844 train_time:150115ms step_avg:100.08ms +step:2000/20000 train_loss:3.1182 train_time:200142ms step_avg:100.07ms +step:2500/20000 train_loss:3.2593 train_time:250128ms step_avg:100.05ms +step:3000/20000 train_loss:2.9743 train_time:300109ms step_avg:100.04ms +step:3500/20000 train_loss:3.1175 train_time:350112ms step_avg:100.03ms +step:4000/20000 train_loss:3.0330 train_time:400056ms step_avg:100.01ms +step:4000/20000 val_loss:3.0169 val_bpb:1.1680 train_time:400123ms step_avg:100.03ms +step:4500/20000 train_loss:3.0503 train_time:449988ms step_avg:100.00ms +step:5000/20000 train_loss:2.8859 train_time:499934ms step_avg:99.99ms +swa:start step:5250 +late_qat:enabled step:5398 scale:0.1499 +step:5500/20000 train_loss:2.8764 train_time:550321ms step_avg:100.06ms +step:5989/20000 val_loss:2.8522 val_bpb:1.1042 train_time:600115ms step_avg:100.20ms +stopping_early: wallclock_cap train_time:600115ms step:5989/20000 +peak memory allocated: 26621 MiB reserved: 26678 MiB +ema:applying EMA weights +DIAGNOSTIC post_ema val_loss:2.8486 val_bpb:1.1028 eval_time:1595ms +ttt:start lr=0.0005 epochs=6 freeze_blocks=2 cosine_decay=True +ttt_adamw:params trainable=35937864 frozen=4112 +ttt_adamw:epoch 1/6 loss:2.9050 time:14.2s +ttt_adamw:epoch 2/6 loss:2.8355 time:28.2s +ttt_adamw:epoch 3/6 loss:2.8127 time:42.2s +ttt_adamw:epoch 4/6 loss:2.7935 time:56.2s +ttt_adamw:epoch 5/6 loss:2.7773 time:70.2s +ttt_adamw:epoch 6/6 loss:2.7661 time:84.2s +ttt_adamw:done elapsed=84.2s +ttt:elapsed=84.2s +DIAGNOSTIC post_ttt val_loss:2.7598 val_bpb:1.0684 eval_time:4728ms +Serialized model: 135395695 bytes +Code size: 116724 bytes +gptq:building non-banked model for Hessian collection... +gptq:collecting hessians from training data (256 batches)... +gptq:collected hessians for 67 layers (training data) +selective_prune: 9839128 ±1 candidates, unpruned=14.43MB target=15.9MB +selective_prune: already fits, no pruning needed +Serialized model int6+brotli+byteshuffle: 15014416 bytes +Total submission size int6+brotli+byteshuffle: 15131140 bytes +final_int6_roundtrip val_loss:2.8269 val_bpb:1.0944 eval_time:6137ms +final_int6_roundtrip_exact val_loss:2.82694867 val_bpb:1.09439993 +final_int6_sliding_window val_loss:2.7897 val_bpb:1.0800 stride:64 eval_time:67203ms +final_int6_sliding_window_exact val_loss:2.78965147 val_bpb:1.07996556 +final_int8_zlib_roundtrip_exact val_loss:2.78965147 val_bpb:1.07996556 diff --git a/records/track_10min_16mb/2026-04-06_SP8192_PreQuantTTT_SDClip_SkipGates/train_seed42.log b/records/track_10min_16mb/2026-04-06_SP8192_PreQuantTTT_SDClip_SkipGates/train_seed42.log new file mode 100644 index 0000000000..ab897f11ac --- /dev/null +++ b/records/track_10min_16mb/2026-04-06_SP8192_PreQuantTTT_SDClip_SkipGates/train_seed42.log @@ -0,0 +1,2559 @@ +from __future__ import annotations +import copy +import glob +import io +import lzma +import math +import os +import random +import subprocess +import sys +import time +import uuid +import zlib +from pathlib import Path +try: + import zstandard + _COMPRESSOR = "zstd" +except ImportError: + _COMPRESSOR = "zlib" +import numpy as np +import sentencepiece as spm +import torch +import torch.distributed as dist +import torch.nn.functional as F +from torch import Tensor, nn +from torch.nn.parallel import DistributedDataParallel as DDP +try: + from flash_attn_interface import flash_attn_func as flash_attn_3_func + HAS_FA3 = True +except ImportError: + HAS_FA3 = False +class Hyperparameters: + data_path = os.environ.get("DATA_PATH", "./data/datasets/fineweb10B_sp8192") + train_files = os.path.join(data_path, "fineweb_train_*.bin") + val_files = os.path.join(data_path, "fineweb_val_*.bin") + tokenizer_path = os.environ.get("TOKENIZER_PATH", "./data/tokenizers/fineweb_8192_bpe.model") + run_id = os.environ.get("RUN_ID", str(uuid.uuid4())) + seed = int(os.environ.get("SEED", 1337)) + val_batch_size = int(os.environ.get("VAL_BATCH_SIZE", 524_288)) + val_loss_every = int(os.environ.get("VAL_LOSS_EVERY", 4000)) + train_log_every = int(os.environ.get("TRAIN_LOG_EVERY", 500)) + iterations = int(os.environ.get("ITERATIONS", 20000)) + warmdown_iters = int(os.environ.get("WARMDOWN_ITERS", 3500)) # legacy; overridden by warmdown_frac if > 0 + warmdown_frac = float(os.environ.get("WARMDOWN_FRAC", 0.667)) # fraction of total time for warmdown + warmup_steps = int(os.environ.get("WARMUP_STEPS", 20)) + train_batch_tokens = int(os.environ.get("TRAIN_BATCH_TOKENS", 786_432)) + train_seq_len = int(os.environ.get("TRAIN_SEQ_LEN", 2048)) + eval_seq_len = int(os.environ.get("EVAL_SEQ_LEN", 2048)) + max_wallclock_seconds = float(os.environ.get("MAX_WALLCLOCK_SECONDS", 600.0)) + qk_gain_init = float(os.environ.get("QK_GAIN_INIT", 4.0)) + vocab_size = int(os.environ.get("VOCAB_SIZE", 8192)) + num_layers = int(os.environ.get("NUM_LAYERS", 11)) + num_kv_heads = int(os.environ.get("NUM_KV_HEADS", 4)) + model_dim = int(os.environ.get("MODEL_DIM", 512)) + num_heads = int(os.environ.get("NUM_HEADS", 8)) + mlp_mult = float(os.environ.get("MLP_MULT", 4.0)) + tie_embeddings = bool(int(os.environ.get("TIE_EMBEDDINGS", "1"))) + rope_base = float(os.environ.get("ROPE_BASE", 10000.0)) + logit_softcap = float(os.environ.get("LOGIT_SOFTCAP", 30.0)) + embed_lr = float(os.environ.get("EMBED_LR", 0.6)) + head_lr = float(os.environ.get("HEAD_LR", 0.008)) + tied_embed_lr = float(os.environ.get("TIED_EMBED_LR", 0.035)) + tied_embed_init_std = float(os.environ.get("TIED_EMBED_INIT_STD", 0.005)) + matrix_lr = float(os.environ.get("MATRIX_LR", 0.025)) + scalar_lr = float(os.environ.get("SCALAR_LR", 0.025)) + muon_momentum = float(os.environ.get("MUON_MOMENTUM", 0.99)) + muon_backend_steps = int(os.environ.get("MUON_BACKEND_STEPS", 5)) + muon_momentum_warmup_start = float(os.environ.get("MUON_MOMENTUM_WARMUP_START", 0.92)) + muon_momentum_warmup_steps = int(os.environ.get("MUON_MOMENTUM_WARMUP_STEPS", 1500)) + beta1 = float(os.environ.get("BETA1", 0.9)) + beta2 = float(os.environ.get("BETA2", 0.95)) + adam_eps = float(os.environ.get("ADAM_EPS", 1e-8)) + grad_clip_norm = float(os.environ.get("GRAD_CLIP_NORM", 0.3)) + eval_stride = int(os.environ.get("EVAL_STRIDE", 64)) + mtp_num_heads = int(os.environ.get("MTP_NUM_HEADS", 0)) + mtp_loss_weight = float(os.environ.get("MTP_LOSS_WEIGHT", 0.2)) + muon_beta2 = float(os.environ.get("MUON_BETA2", 0.95)) + swa_enabled = bool(int(os.environ.get("SWA_ENABLED", "1"))) + swa_every = int(os.environ.get("SWA_EVERY", 50)) + lawa_enabled = bool(int(os.environ.get("LAWA_ENABLED", "0"))) + lawa_k = int(os.environ.get("LAWA_K", 10)) + lawa_freq = int(os.environ.get("LAWA_FREQ", 100)) + muon_wd = float(os.environ.get("MUON_WD", 0.085)) + adam_wd = float(os.environ.get("ADAM_WD", 0.02)) + embed_wd = float(os.environ.get("EMBED_WD", 0.085)) + qat_enabled = bool(int(os.environ.get("QAT_ENABLED", "0"))) + bigram_vocab_size = int(os.environ.get("BIGRAM_VOCAB_SIZE", 0)) # disabled: SP8192 replaces BigramHash + bigram_dim = int(os.environ.get("BIGRAM_DIM", 128)) + trigram_enabled = bool(int(os.environ.get("TRIGRAM", "0"))) # TrigramHash (off by default, risky) + xsa_last_n = int(os.environ.get("XSA_LAST_N", 11)) # XSA on ALL layers (our novel contribution) + rope_dims = int(os.environ.get("ROPE_DIMS", 16)) + ln_scale = bool(int(os.environ.get("LN_SCALE", "1"))) + dtg_enabled = bool(int(os.environ.get("DTG_ENABLED", "0"))) + late_qat_threshold = float(os.environ.get("LATE_QAT_THRESHOLD", 0.15)) + ve_enabled = bool(int(os.environ.get("VE_ENABLED", "0"))) + ve_dim = int(os.environ.get("VE_DIM", 128)) + ve_layers = os.environ.get("VE_LAYERS", "9,10") + gated_attention = bool(int(os.environ.get("GATED_ATTENTION", "0"))) + value_residual = bool(int(os.environ.get("VALUE_RESIDUAL", "0"))) # VRL with sigmoid gates (off by default, risky) + skip_gates_enabled = bool(int(os.environ.get("SKIP_GATES_ENABLED", "1"))) # learned sigmoid gates on U-Net skips + # GPTQ calibration + gptq_calib_batches = int(os.environ.get("GPTQ_CALIB_BATCHES", 256)) + gptq_block_size = int(os.environ.get("GPTQ_BLOCK_SIZE", 128)) + # SDClip: clip threshold = k * std(row) instead of percentile search + matrix_bits = int(os.environ.get("MATRIX_BITS", 6)) + embed_bits = int(os.environ.get("EMBED_BITS", 8)) + matrix_clip_sigmas = float(os.environ.get("MATRIX_CLIP_SIGMAS", 12.85)) # k for int6 matrices + embed_clip_sigmas = float(os.environ.get("EMBED_CLIP_SIGMAS", 20.0)) # k for int8 embeddings + # TTT: AdamW test-time training (pre-quantization, on EMA model) + ttt_enabled = bool(int(os.environ.get("TTT_ENABLED", "1"))) + ttt_lr = float(os.environ.get("TTT_LR", 0.0005)) + ttt_epochs = int(os.environ.get("TTT_EPOCHS", 6)) + ttt_batch_seqs = int(os.environ.get("TTT_BATCH_SEQS", 32)) + ttt_freeze_blocks = int(os.environ.get("TTT_FREEZE_BLOCKS", 2)) + ttt_grad_clip = float(os.environ.get("TTT_GRAD_CLIP", 1.0)) + ttt_cosine_decay = bool(int(os.environ.get("TTT_COSINE_DECAY", "1"))) + # Causal SLOT: per-batch delta optimization at last hidden layer (post-quantization) + slot_enabled = bool(int(os.environ.get("SLOT_ENABLED", "0"))) + slot_lr = float(os.environ.get("SLOT_LR", 0.005)) + slot_steps = int(os.environ.get("SLOT_STEPS", 8)) + +# --- Batched Newton-Schulz orthogonalization --- + +def zeropower_via_newtonschulz5(G: Tensor, steps: int = 5, eps: float = 1e-7) -> Tensor: + """Batched Newton-Schulz orthogonalization. G: (B,M,N) or (M,N).""" + a, b, c = (3.4445, -4.7750, 2.0315) + was_2d = G.ndim == 2 + if was_2d: + G = G.unsqueeze(0) + X = G.bfloat16() + transposed = X.size(-2) > X.size(-1) + if transposed: + X = X.mT + X = X / (X.norm(dim=(-2, -1), keepdim=True) + eps) + for _ in range(steps): + A = X @ X.mT + B = b * A + c * (A @ A) + X = a * X + B @ X + if transposed: + X = X.mT + if was_2d: + X = X.squeeze(0) + return X + +# --- Parallel Muon optimizer --- + +class Muon(torch.optim.Optimizer): + """Parallel Muon: post-backward reduce-scatter -> local NS5 -> all-gather. + + No DDP for bank params. After backward, this optimizer: + 1. Launches async reduce-scatter for all banks (biggest first) + 2. Returns control so Adam can step on small params while RS is in-flight + 3. Waits for each RS, runs local NS5 on the shard, launches async all-gather + 4. Each all-gather overlaps with next bank's NS5 + """ + def __init__(self, params, lr: float, momentum: float, backend_steps: int, + nesterov: bool = True, weight_decay: float = 0.0): + super().__init__( + params, + dict(lr=lr, momentum=momentum, backend_steps=backend_steps, + nesterov=nesterov, weight_decay=weight_decay), + ) + self._built = False + + def _build(self): + self._distributed = dist.is_available() and dist.is_initialized() + self._world_size = dist.get_world_size() if self._distributed else 1 + self._rank = dist.get_rank() if self._distributed else 0 + ws = self._world_size + + self._bank_meta = [] + for group in self.param_groups: + for p in group["params"]: + B = p.shape[0] + padded_B = ((B + ws - 1) // ws) * ws + shard_B = padded_B // ws + tail = p.shape[1:] + dev = p.device + self._bank_meta.append({ + 'p': p, + 'B': B, + 'padded_grad': torch.zeros(padded_B, *tail, device=dev, dtype=torch.bfloat16), + 'shard': torch.zeros(shard_B, *tail, device=dev, dtype=torch.bfloat16), + 'shard_mom': torch.zeros(shard_B, *tail, device=dev, dtype=torch.bfloat16), + 'full_update': torch.zeros(padded_B, *tail, device=dev, dtype=torch.bfloat16), + 'scale': max(1, p.shape[-2] / p.shape[-1]) ** 0.5, + }) + # Sort by size descending -- launch biggest reduce-scatters first + self._bank_meta.sort(key=lambda m: -m['p'].numel()) + self._built = True + + def launch_reduce_scatters(self): + """Phase 1: launch async reduce-scatter for all banks. Call right after backward.""" + if not self._built: + self._build() + if not self._distributed: + return + self._rs_futures = [] + for m in self._bank_meta: + p = m['p'] + if p.grad is None: + self._rs_futures.append(None) + continue + pg = m['padded_grad'] + pg[:m['B']].copy_(p.grad.bfloat16()) + if pg.shape[0] > m['B']: + pg[m['B']:].zero_() + fut = dist.reduce_scatter_tensor(m['shard'], pg, op=dist.ReduceOp.AVG, async_op=True) + self._rs_futures.append(fut) + + @torch.no_grad() + def step(self, closure=None): + """Phase 3: wait for RS, local NS5, all-gather. Call AFTER Adam steps.""" + loss = None + if closure is not None: + with torch.enable_grad(): + loss = closure() + + if not self._built: + self._build() + + for group in self.param_groups: + lr = group["lr"] + momentum = group["momentum"] + backend_steps = group["backend_steps"] + nesterov = group["nesterov"] + wd = group.get("weight_decay", 0.0) + + prev_ag_handle = None + prev_m = None + + sharded = self._distributed and hasattr(self, '_rs_futures') + + for i, m in enumerate(self._bank_meta): + p = m['p'] + if p.grad is None: + continue + + if prev_ag_handle is not None: + prev_ag_handle.wait() + pp = prev_m['p'] + upd = prev_m['full_update'][:prev_m['B']] + if wd > 0.0: + pp.data.mul_(1.0 - lr * wd) + pp.add_(upd.to(dtype=pp.dtype), alpha=-lr * prev_m['scale']) + + if sharded and self._rs_futures[i] is not None: + self._rs_futures[i].wait() + g = m['shard'] + buf = m['shard_mom'] + else: + g = p.grad.bfloat16() + state = self.state[p] + if "momentum_buffer" not in state: + state["momentum_buffer"] = torch.zeros_like(g) + buf = state["momentum_buffer"] + + buf.mul_(momentum).add_(g) + if nesterov: + update = g.add(buf, alpha=momentum) + else: + update = buf + + update = zeropower_via_newtonschulz5(update, steps=backend_steps) + + if sharded: + prev_ag_handle = dist.all_gather_into_tensor( + m['full_update'], update, async_op=True) + prev_m = m + else: + if wd > 0.0: + p.data.mul_(1.0 - lr * wd) + p.add_(update.to(dtype=p.dtype), alpha=-lr * m['scale']) + + if prev_ag_handle is not None: + prev_ag_handle.wait() + pp = prev_m['p'] + upd = prev_m['full_update'][:prev_m['B']] + if wd > 0.0: + pp.data.mul_(1.0 - lr * wd) + pp.add_(upd.to(dtype=pp.dtype), alpha=-lr * prev_m['scale']) + + if hasattr(self, '_rs_futures'): + del self._rs_futures + + return loss + +# --- Tokenizer evaluation helpers --- + +def build_sentencepiece_luts( + sp: spm.SentencePieceProcessor, vocab_size: int, device: torch.device +) -> tuple[Tensor, Tensor, Tensor]: + sp_vocab_size = int(sp.vocab_size()) + table_size = max(sp_vocab_size, vocab_size) + base_bytes_np = np.zeros((table_size,), dtype=np.int16) + has_leading_space_np = np.zeros((table_size,), dtype=np.bool_) + is_boundary_token_np = np.ones((table_size,), dtype=np.bool_) + for token_id in range(sp_vocab_size): + if sp.is_control(token_id) or sp.is_unknown(token_id) or sp.is_unused(token_id): + continue + is_boundary_token_np[token_id] = False + if sp.is_byte(token_id): + base_bytes_np[token_id] = 1 + continue + piece = sp.id_to_piece(token_id) + if piece.startswith("\u2581"): + has_leading_space_np[token_id] = True + piece = piece[1:] + base_bytes_np[token_id] = len(piece.encode("utf-8")) + return ( + torch.tensor(base_bytes_np, dtype=torch.int16, device=device), + torch.tensor(has_leading_space_np, dtype=torch.bool, device=device), + torch.tensor(is_boundary_token_np, dtype=torch.bool, device=device), + ) +def load_validation_tokens(pattern: str, seq_len: int) -> Tensor: + files = [Path(p) for p in sorted(glob.glob(pattern))] + if not files: + raise FileNotFoundError(f"No files found for pattern: {pattern}") + tokens = torch.cat([load_data_shard(file) for file in files]).contiguous() + usable = ((tokens.numel() - 1) // seq_len) * seq_len + if usable <= 0: + raise ValueError(f"Validation split is too short for TRAIN_SEQ_LEN={seq_len}") + return tokens[: usable + 1] +def eval_val( + args: Hyperparameters, + model: nn.Module, + rank: int, + world_size: int, + device: torch.device, + grad_accum_steps: int, + val_tokens: Tensor, + base_bytes_lut: Tensor, + has_leading_space_lut: Tensor, + is_boundary_token_lut: Tensor, + eval_seq_len: int | None = None, +) -> tuple[float, float]: + seq_len = eval_seq_len or args.train_seq_len + local_batch_tokens = args.val_batch_size // (world_size * grad_accum_steps) + if local_batch_tokens < seq_len: + raise ValueError( + "VAL_BATCH_SIZE must provide at least one sequence per rank; " + f"got VAL_BATCH_SIZE={args.val_batch_size}, WORLD_SIZE={world_size}, " + f"GRAD_ACCUM_STEPS={grad_accum_steps}, seq_len={seq_len}" + ) + local_batch_seqs = local_batch_tokens // seq_len + total_seqs = (val_tokens.numel() - 1) // seq_len + seq_start = (total_seqs * rank) // world_size + seq_end = (total_seqs * (rank + 1)) // world_size + val_loss_sum = torch.zeros((), device=device, dtype=torch.float64) + val_token_count = torch.zeros((), device=device, dtype=torch.float64) + val_byte_count = torch.zeros((), device=device, dtype=torch.float64) + model.eval() + with torch.inference_mode(): + for batch_seq_start in range(seq_start, seq_end, local_batch_seqs): + batch_seq_end = min(batch_seq_start + local_batch_seqs, seq_end) + raw_start = batch_seq_start * seq_len + raw_end = batch_seq_end * seq_len + 1 + local = val_tokens[raw_start:raw_end].to(device=device, dtype=torch.int64, non_blocking=True) + x = local[:-1].reshape(-1, seq_len) + y = local[1:].reshape(-1, seq_len) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + batch_loss = model(x, y).detach() + batch_token_count = float(y.numel()) + val_loss_sum += batch_loss.to(torch.float64) * batch_token_count + val_token_count += batch_token_count + prev_ids = x.reshape(-1) + tgt_ids = y.reshape(-1) + token_bytes = base_bytes_lut[tgt_ids].to(dtype=torch.int16) + token_bytes += (has_leading_space_lut[tgt_ids] & ~is_boundary_token_lut[prev_ids]).to(dtype=torch.int16) + val_byte_count += token_bytes.to(torch.float64).sum() + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(val_loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(val_token_count, op=dist.ReduceOp.SUM) + dist.all_reduce(val_byte_count, op=dist.ReduceOp.SUM) + val_loss = val_loss_sum / val_token_count + bits_per_token = val_loss.item() / math.log(2.0) + tokens_per_byte = val_token_count.item() / val_byte_count.item() + model.train() + return float(val_loss.item()), float(bits_per_token * tokens_per_byte) + +# --- Quantization helpers --- + +CONTROL_TENSOR_NAME_PATTERNS = tuple( + pattern + for pattern in os.environ.get( + "CONTROL_TENSOR_NAME_PATTERNS", + "attn_scale,attn_scales,mlp_scale,mlp_scales,resid_mix,resid_mixes,q_gain,skip_weight,skip_weights,skip_gates,smear,dtg_gate,ve_layer_scales,ve_shared.scale,attn_gate,vr_lambda", + ).split(",") + if pattern +) +INT8_KEEP_FLOAT_FP32_NAME_PATTERNS = tuple( + pattern + for pattern in os.environ.get( + "INT8_KEEP_FLOAT_FP32_NAME_PATTERNS", + ",".join(CONTROL_TENSOR_NAME_PATTERNS), + ).split(",") + if pattern +) +INT8_KEEP_FLOAT_MAX_NUMEL = 65_536 +INT8_KEEP_FLOAT_STORE_DTYPE = torch.float16 +INT8_PER_ROW_SCALE_DTYPE = torch.float16 +INT8_CLIP_PERCENTILE = 99.99984 +INT8_CLIP_Q = INT8_CLIP_PERCENTILE / 100.0 +def tensor_nbytes(t: Tensor) -> int: + return int(t.numel()) * int(t.element_size()) +def keep_float_tensor(name: str, t: Tensor, passthrough_orig_dtypes: dict[str, str]) -> Tensor: + if any(pattern in name for pattern in INT8_KEEP_FLOAT_FP32_NAME_PATTERNS): + return t.float().contiguous() + if t.dtype in {torch.float32, torch.bfloat16}: + passthrough_orig_dtypes[name] = str(t.dtype).removeprefix("torch.") + return t.to(dtype=INT8_KEEP_FLOAT_STORE_DTYPE).contiguous() + return t +def quantize_float_tensor(t: Tensor) -> tuple[Tensor, Tensor]: + t32 = t.float() + if t32.ndim == 2: + clip_abs = ( + torch.quantile(t32.abs(), INT8_CLIP_Q, dim=1) + if t32.numel() + else torch.empty((t32.shape[0],), dtype=torch.float32) + ) + clipped = torch.maximum(torch.minimum(t32, clip_abs[:, None]), -clip_abs[:, None]) + scale = (clip_abs / 127.0).clamp_min(1.0 / 127.0) + q = torch.clamp(torch.round(clipped / scale[:, None]), -127, 127).to(torch.int8).contiguous() + return q, scale.to(dtype=INT8_PER_ROW_SCALE_DTYPE).contiguous() + clip_abs = float(torch.quantile(t32.abs().flatten(), INT8_CLIP_Q).item()) if t32.numel() else 0.0 + scale = torch.tensor(clip_abs / 127.0 if clip_abs > 0 else 1.0, dtype=torch.float32) + q = torch.clamp(torch.round(torch.clamp(t32, -clip_abs, clip_abs) / scale), -127, 127).to(torch.int8).contiguous() + return q, scale +def quantize_state_dict_int8(state_dict: dict[str, Tensor]): + quantized: dict[str, Tensor] = {} + scales: dict[str, Tensor] = {} + dtypes: dict[str, str] = {} + passthrough: dict[str, Tensor] = {} + passthrough_orig_dtypes: dict[str, str] = {} + qmeta: dict[str, dict[str, object]] = {} + stats = dict.fromkeys( + ("param_count", "num_tensors", "num_float_tensors", "num_nonfloat_tensors", "baseline_tensor_bytes", "int8_payload_bytes"), + 0, + ) + for name, tensor in state_dict.items(): + t = tensor.detach().to("cpu").contiguous() + stats["param_count"] += int(t.numel()) + stats["num_tensors"] += 1 + stats["baseline_tensor_bytes"] += tensor_nbytes(t) + if not t.is_floating_point(): + stats["num_nonfloat_tensors"] += 1 + passthrough[name] = t + stats["int8_payload_bytes"] += tensor_nbytes(t) + continue + if t.numel() <= INT8_KEEP_FLOAT_MAX_NUMEL: + kept = keep_float_tensor(name, t, passthrough_orig_dtypes) + passthrough[name] = kept + stats["int8_payload_bytes"] += tensor_nbytes(kept) + continue + stats["num_float_tensors"] += 1 + q, s = quantize_float_tensor(t) + if s.ndim > 0: + qmeta[name] = {"scheme": "per_row", "axis": 0} + quantized[name] = q + scales[name] = s + dtypes[name] = str(t.dtype).removeprefix("torch.") + stats["int8_payload_bytes"] += tensor_nbytes(q) + tensor_nbytes(s) + obj: dict[str, object] = { + "__quant_format__": "int8_clean_per_row_v1", + "quantized": quantized, + "scales": scales, + "dtypes": dtypes, + "passthrough": passthrough, + } + if qmeta: + obj["qmeta"] = qmeta + if passthrough_orig_dtypes: + obj["passthrough_orig_dtypes"] = passthrough_orig_dtypes + return obj, stats +def dequantize_state_dict_int8(obj: dict[str, object]) -> dict[str, Tensor]: + out: dict[str, Tensor] = {} + qmeta = obj.get("qmeta", {}) + passthrough_orig_dtypes = obj.get("passthrough_orig_dtypes", {}) + for name, q in obj["quantized"].items(): + dtype = getattr(torch, obj["dtypes"][name]) + s = obj["scales"][name] + if qmeta.get(name, {}).get("scheme") == "per_row" or s.ndim > 0: + s = s.to(dtype=torch.float32) + out[name] = (q.float() * s.view(q.shape[0], *([1] * (q.ndim - 1)))).to(dtype=dtype).contiguous() + else: + scale = float(s.item()) + out[name] = (q.float() * scale).to(dtype=dtype).contiguous() + for name, t in obj["passthrough"].items(): + out_t = t.detach().to("cpu").contiguous() + orig_dtype = passthrough_orig_dtypes.get(name) + if isinstance(orig_dtype, str): + out_t = out_t.to(dtype=getattr(torch, orig_dtype)).contiguous() + out[name] = out_t + return out + +# --- Data loading --- + +def load_data_shard(file: Path) -> Tensor: + header_bytes = 256 * np.dtype(" None: + self.file_idx = (self.file_idx + 1) % len(self.files) + self.tokens = load_data_shard(self.files[self.file_idx]) + self.pos = 0 + def take(self, n: int) -> Tensor: + chunks: list[Tensor] = [] + remaining = n + while remaining > 0: + avail = self.tokens.numel() - self.pos + if avail <= 0: + self._advance_file() + continue + k = min(remaining, avail) + chunks.append(self.tokens[self.pos : self.pos + k]) + self.pos += k + remaining -= k + return chunks[0] if len(chunks) == 1 else torch.cat(chunks) +class DistributedTokenLoader: + def __init__(self, pattern: str, rank: int, world_size: int, device: torch.device): + self.rank = rank + self.world_size = world_size + self.device = device + self.stream = TokenStream(pattern) + def next_batch(self, global_tokens: int, seq_len: int, grad_accum_steps: int) -> tuple[Tensor, Tensor]: + local_tokens = global_tokens // (self.world_size * grad_accum_steps) + per_rank_span = local_tokens + 1 + chunk = self.stream.take(per_rank_span * self.world_size) + start = self.rank * per_rank_span + local = chunk[start : start + per_rank_span].to(dtype=torch.int64) + x = local[:-1].reshape(-1, seq_len) + y = local[1:].reshape(-1, seq_len) + return x.to(self.device, non_blocking=True), y.to(self.device, non_blocking=True) + +# --- Transformer modules --- + +class RMSNorm(nn.Module): + def __init__(self, eps: float | None = None): + super().__init__() + self.eps = eps + def forward(self, x: Tensor) -> Tensor: + return F.rms_norm(x, (x.size(-1),), eps=self.eps) +class CastedLinear(nn.Linear): + _qat_enabled: bool = False + def forward(self, x: Tensor) -> Tensor: + w = self.weight.to(x.dtype) + if CastedLinear._qat_enabled and self.training and w.ndim == 2: + with torch.no_grad(): + w32 = self.weight.float() + row_max = w32.abs().amax(dim=1) + scale = (row_max / 31.0).clamp_min(1.0 / 31.0) + w_q = (torch.clamp(torch.round(w32 / scale[:, None]), -32, 31) * scale[:, None]).to(x.dtype) + w = w + (w_q - w).detach() + bias = self.bias.to(x.dtype) if self.bias is not None else None + return F.linear(x, w, bias) +def restore_low_dim_params_to_fp32(module: nn.Module) -> None: + with torch.no_grad(): + for name, param in module.named_parameters(): + if (param.ndim < 2 or any(pattern in name for pattern in CONTROL_TENSOR_NAME_PATTERNS)) and param.dtype != torch.float32: + param.data = param.data.float() +class Rotary(nn.Module): + def __init__(self, dim: int, base: float = 10000.0, train_seq_len: int = 1024, rope_dims: int = 0): + super().__init__() + self.dim = dim + self.base = base + self.train_seq_len = train_seq_len + self.rope_dims = rope_dims if rope_dims > 0 else dim + inv_freq = 1.0 / (base ** (torch.arange(0, self.rope_dims, 2, dtype=torch.float32) / self.rope_dims)) + self.register_buffer("inv_freq", inv_freq, persistent=False) + self._seq_len_cached = 0 + self._cos_cached: Tensor | None = None + self._sin_cached: Tensor | None = None + def forward(self, seq_len: int, device: torch.device, dtype: torch.dtype) -> tuple[Tensor, Tensor]: + if ( + self._cos_cached is None + or self._sin_cached is None + or self._seq_len_cached != seq_len + or self._cos_cached.device != device + ): + rd = self.rope_dims + if seq_len > self.train_seq_len: + scale = seq_len / self.train_seq_len + new_base = self.base * (scale ** (rd / (rd - 2))) + inv_freq = 1.0 / (new_base ** (torch.arange(0, rd, 2, dtype=torch.float32, device=device) / rd)) + else: + inv_freq = self.inv_freq.to(device) + t = torch.arange(seq_len, device=device, dtype=inv_freq.dtype) + freqs = torch.outer(t, inv_freq) + self._cos_cached = freqs.cos()[None, :, None, :] + self._sin_cached = freqs.sin()[None, :, None, :] + self._seq_len_cached = seq_len + return self._cos_cached.to(dtype=dtype), self._sin_cached.to(dtype=dtype) +def apply_rotary_emb(x: Tensor, cos: Tensor, sin: Tensor, rope_dims: int = 0) -> Tensor: + if rope_dims > 0 and rope_dims < x.size(-1): + x_rope, x_pass = x[..., :rope_dims], x[..., rope_dims:] + half = rope_dims // 2 + x1, x2 = x_rope[..., :half], x_rope[..., half:] + x_rope = torch.cat((x1 * cos + x2 * sin, x1 * (-sin) + x2 * cos), dim=-1) + return torch.cat((x_rope, x_pass), dim=-1) + half = x.size(-1) // 2 + x1, x2 = x[..., :half], x[..., half:] + return torch.cat((x1 * cos + x2 * sin, x1 * (-sin) + x2 * cos), dim=-1) + +class CausalSelfAttention(nn.Module): + def __init__( + self, + dim: int, + num_heads: int, + num_kv_heads: int, + rope_base: float, + qk_gain_init: float, + gated_attention: bool = False, + value_residual: bool = False, + ): + super().__init__() + if dim % num_heads != 0: + raise ValueError("model_dim must be divisible by num_heads") + if num_heads % num_kv_heads != 0: + raise ValueError("num_heads must be divisible by num_kv_heads") + self.num_heads = num_heads + self.num_kv_heads = num_kv_heads + self.head_dim = dim // num_heads + if self.head_dim % 2 != 0: + raise ValueError("head_dim must be even for RoPE") + # No CastedLinear -- weights come from banks + self.q_gain = nn.Parameter(torch.full((num_heads,), qk_gain_init, dtype=torch.float32)) + self.rope_dims = 0 # set by GPT.__init__ for partial RoPE + self.rotary = Rotary(self.head_dim, base=rope_base, train_seq_len=1024) + self.use_xsa = False # set by GPT.__init__ for deep layers only + # Gated attention and value residual (non-banked small params) + self.gated_attention = gated_attention + if gated_attention: + self.attn_gate = nn.Linear(dim, num_heads, bias=True) + nn.init.zeros_(self.attn_gate.weight) + nn.init.constant_(self.attn_gate.bias, 4.0) + self.value_residual = value_residual + if value_residual: + self.vrl_alpha = nn.Parameter(torch.zeros(1, dtype=torch.float32)) # sigmoid gate (PR #569 style) + def _xsa_efficient(self, y: Tensor, v: Tensor) -> Tensor: + """Efficient XSA: subtract self-value projection via GQA-aware reshape (no repeat_interleave). + y: [B, T, H, D], v: [B, T, Hkv, D]. H must be divisible by Hkv.""" + B, T, H, D = y.shape + Hkv = v.size(-2) + group = H // Hkv + y_g = y.reshape(B, T, Hkv, group, D) # [B, T, Hkv, group, D] + vn = F.normalize(v, dim=-1).unsqueeze(-2) # [B, T, Hkv, 1, D] -- broadcast ready + proj = (y_g * vn).sum(dim=-1, keepdim=True) * vn + return (y_g - proj).reshape(B, T, H, D) + def forward(self, x: Tensor, q_w: Tensor, k_w: Tensor, v_w: Tensor, out_w: Tensor, v_embed: Tensor | None = None, v0: Tensor | None = None) -> tuple[Tensor, Tensor | None]: + bsz, seqlen, dim = x.shape + q = F.linear(x, q_w.to(x.dtype)).reshape(bsz, seqlen, self.num_heads, self.head_dim) + k = F.linear(x, k_w.to(x.dtype)).reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + v = F.linear(x, v_w.to(x.dtype)) + if v_embed is not None: + v = v + v_embed + v = v.reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + raw_v = v if self.value_residual else None + if self.value_residual and v0 is not None: + alpha = torch.sigmoid(self.vrl_alpha.to(dtype=v.dtype)) + v = v + alpha * v0 # sigmoid-gated residual (PR #569 style) + q = F.rms_norm(q, (q.size(-1),)) + k = F.rms_norm(k, (k.size(-1),)) + cos, sin = self.rotary(seqlen, x.device, q.dtype) + q = apply_rotary_emb(q, cos, sin, self.rope_dims) + k = apply_rotary_emb(k, cos, sin, self.rope_dims) + q = q * self.q_gain.to(dtype=q.dtype)[None, None, :, None] + if HAS_FA3: + y = flash_attn_3_func(q, k, v, causal=True) + else: + q2, k2, v2 = q.transpose(1, 2), k.transpose(1, 2), v.transpose(1, 2) + if k2.size(1) != q2.size(1): + reps = q2.size(1) // k2.size(1) + k2 = k2.repeat_interleave(reps, dim=1) + v2 = v2.repeat_interleave(reps, dim=1) + y = F.scaled_dot_product_attention(q2, k2, v2, is_causal=True).transpose(1, 2) + if self.use_xsa: + y = self._xsa_efficient(y, v) + if self.gated_attention: + # gate shape: (bsz, seqlen, num_heads) -> (bsz, seqlen, num_heads, 1) for B,T,H,D layout + gate = torch.sigmoid(self.attn_gate(x)).unsqueeze(-1) + y = y * gate + y = y.reshape(bsz, seqlen, dim) + return F.linear(y, out_w.to(x.dtype)), raw_v + +class SmearGate(nn.Module): + def __init__(self, dim: int): + super().__init__() + self.gate = nn.Parameter(torch.zeros(dim, dtype=torch.float32)) + def forward(self, x: Tensor) -> Tensor: + g = torch.sigmoid(self.gate.to(dtype=x.dtype))[None, None, :] + x_prev = torch.cat([torch.zeros_like(x[:, :1]), x[:, :-1]], dim=1) + return (1 - g) * x + g * x_prev + +class BigramHashEmbedding(nn.Module): + def __init__(self, bigram_vocab_size: int, bigram_dim: int, model_dim: int, trigram: bool = False): + super().__init__() + self.bigram_vocab_size = bigram_vocab_size + self._trigram = trigram + self.embed = nn.Embedding(bigram_vocab_size, bigram_dim) + nn.init.zeros_(self.embed.weight) + self.proj = CastedLinear(bigram_dim, model_dim, bias=False) if bigram_dim != model_dim else None + if self.proj is not None: + nn.init.zeros_(self.proj.weight) + self.scale = nn.Parameter(torch.tensor(0.05, dtype=torch.float32)) + def bigram_hash(self, tokens: Tensor) -> Tensor: + t = tokens.to(torch.int32) + mod = self.bigram_vocab_size - 1 + out = torch.empty_like(t) + out[..., 0] = mod + out[..., 1:] = torch.bitwise_xor(36313 * t[..., 1:], 27191 * t[..., :-1]) % mod + return out.long() + def trigram_hash(self, tokens: Tensor) -> Tensor: + """Hash (t-2, t-1, t) trigrams into same embedding table. Zero extra params.""" + t = tokens.to(torch.int32) + mod = self.bigram_vocab_size - 1 + out = torch.empty_like(t) + out[..., :2] = mod + out[..., 2:] = (36313 * t[..., 2:] ^ 27191 * t[..., 1:-1] ^ 51497 * t[..., :-2]) % mod + return out.long() + def forward(self, token_ids: Tensor) -> Tensor: + h = self.embed(self.bigram_hash(token_ids)) + if self._trigram: + h = h + self.embed(self.trigram_hash(token_ids)) + if self.proj is not None: + h = self.proj(h) + return h * self.scale.to(dtype=h.dtype) + +class ValueEmbedding(nn.Module): + """Reinject token identity into attention values at specific layers. + Each table maps vocab tokens to a low-dim embedding, projected to model_dim.""" + def __init__(self, vocab_size: int, ve_dim: int, model_dim: int): + super().__init__() + self.embed = nn.Embedding(vocab_size, ve_dim) + nn.init.normal_(self.embed.weight, std=0.01) + self.proj = CastedLinear(ve_dim, model_dim, bias=False) if ve_dim != model_dim else None + if self.proj is not None: + nn.init.zeros_(self.proj.weight) + self.scale = nn.Parameter(torch.tensor(0.1, dtype=torch.float32)) + def forward(self, token_ids: Tensor) -> Tensor: + h = self.embed(token_ids) + if self.proj is not None: + h = self.proj(h) + return h * self.scale.to(dtype=h.dtype) + +class MLP(nn.Module): + def __init__(self, dim: int, mlp_mult: int): + super().__init__() + # No CastedLinear -- weights come from banks + def forward(self, x: Tensor, up_w: Tensor, down_w: Tensor) -> Tensor: + x = F.leaky_relu(F.linear(x, up_w.to(x.dtype)), negative_slope=0.5) + return F.linear(x.square(), down_w.to(x.dtype)) + +class Block(nn.Module): + def __init__( + self, + dim: int, + num_heads: int, + num_kv_heads: int, + mlp_mult: int, + rope_base: float, + qk_gain_init: float, + layer_idx: int = 0, + ln_scale: bool = False, + dtg: bool = False, + gated_attention: bool = False, + value_residual: bool = False, + ): + super().__init__() + self.attn_norm = RMSNorm() + self.mlp_norm = RMSNorm() + self.attn = CausalSelfAttention(dim, num_heads, num_kv_heads, rope_base, qk_gain_init, + gated_attention=gated_attention, value_residual=value_residual) + self.mlp = MLP(dim, mlp_mult) + self.attn_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.mlp_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.resid_mix = nn.Parameter(torch.stack((torch.ones(dim), torch.zeros(dim))).float()) + self.ln_scale_factor = 1.0 / math.sqrt(layer_idx + 1) if ln_scale else 1.0 + if dtg: + self.dtg_gate = nn.Linear(dim, 1, bias=True) + nn.init.zeros_(self.dtg_gate.weight) + nn.init.constant_(self.dtg_gate.bias, 2.0) + else: + self.dtg_gate = None + def forward(self, x: Tensor, x0: Tensor, q_w: Tensor, k_w: Tensor, v_w: Tensor, out_w: Tensor, up_w: Tensor, down_w: Tensor, v_embed: Tensor | None = None, v0: Tensor | None = None) -> tuple[Tensor, Tensor | None]: + mix = self.resid_mix.to(dtype=x.dtype) + x_in = mix[0][None, None, :] * x + mix[1][None, None, :] * x0 + attn_out, raw_v = self.attn(self.attn_norm(x_in) * self.ln_scale_factor, q_w, k_w, v_w, out_w, v_embed=v_embed, v0=v0) + x_out = x_in + self.attn_scale.to(dtype=x_in.dtype)[None, None, :] * attn_out + x_out = x_out + self.mlp_scale.to(dtype=x_out.dtype)[None, None, :] * self.mlp(self.mlp_norm(x_out) * self.ln_scale_factor, up_w, down_w) + if self.dtg_gate is not None: + gate = torch.sigmoid(self.dtg_gate(x_in.detach())) + x_out = x_in + gate * (x_out - x_in) + return x_out, raw_v + +class GPT(nn.Module): + def __init__( + self, + vocab_size: int, + num_layers: int, + model_dim: int, + num_heads: int, + num_kv_heads: int, + mlp_mult: int, + tie_embeddings: bool, + tied_embed_init_std: float, + logit_softcap: float, + rope_base: float, + qk_gain_init: float, + mtp_num_heads: int = 0, + mtp_loss_weight: float = 0.1, + bigram_vocab_size: int = 0, + bigram_dim: int = 128, + xsa_last_n: int = 0, + rope_dims: int = 0, + ln_scale: bool = False, + dtg: bool = False, + ve_enabled: bool = False, + ve_dim: int = 128, + ve_layers: str = "9,10", + gated_attention: bool = False, + value_residual: bool = False, + skip_gates_enabled: bool = True, + ): + super().__init__() + self._ve_target_dim = num_kv_heads * (model_dim // num_heads) # kv_dim for value projection + if logit_softcap <= 0.0: + raise ValueError(f"logit_softcap must be positive, got {logit_softcap}") + self.tie_embeddings = tie_embeddings + self.tied_embed_init_std = tied_embed_init_std + self.logit_softcap = logit_softcap + self.value_residual = value_residual + self.mtp_num_heads = mtp_num_heads + self.mtp_loss_weight = mtp_loss_weight + self.tok_emb = nn.Embedding(vocab_size, model_dim) + self.bigram = BigramHashEmbedding(bigram_vocab_size, bigram_dim, model_dim, trigram=bool(int(os.environ.get("TRIGRAM", "0")))) if bigram_vocab_size > 0 else None + self.smear = SmearGate(model_dim) + self.num_encoder_layers = num_layers // 2 + self.num_decoder_layers = num_layers - self.num_encoder_layers + self.num_skip_weights = min(self.num_encoder_layers, self.num_decoder_layers) + self.skip_weights = nn.Parameter(torch.ones(self.num_skip_weights, model_dim, dtype=torch.float32)) + self.skip_gates = nn.Parameter(torch.zeros(self.num_skip_weights, model_dim, dtype=torch.float32)) if skip_gates_enabled else None + # Parameter banks: contiguous 3D tensors for batched optimizer + head_dim = model_dim // num_heads + kv_dim = num_kv_heads * head_dim + mlp_dim = int(mlp_mult * model_dim) + self.num_layers = num_layers + self.qo_bank = nn.Parameter(torch.empty(2 * num_layers, model_dim, model_dim)) + self.kv_bank = nn.Parameter(torch.empty(2 * num_layers, kv_dim, model_dim)) + self.mlp_up_bank = nn.Parameter(torch.empty(num_layers, mlp_dim, model_dim)) + self.mlp_down_bank = nn.Parameter(torch.empty(num_layers, model_dim, mlp_dim)) + self.blocks = nn.ModuleList( + [ + Block( + model_dim, + num_heads, + num_kv_heads, + mlp_mult, + rope_base, + qk_gain_init, + layer_idx=i, + ln_scale=ln_scale, + dtg=dtg, + gated_attention=gated_attention, + value_residual=value_residual, + ) + for i in range(num_layers) + ] + ) + if rope_dims > 0: + head_dim = model_dim // num_heads + for block in self.blocks: + block.attn.rope_dims = rope_dims + block.attn.rotary = Rotary(head_dim, base=rope_base, train_seq_len=1024, rope_dims=rope_dims) + self.ve_layer_indices = [int(x) for x in ve_layers.split(",") if x.strip()] if ve_enabled else [] + kv_dim_ve = self._ve_target_dim + if self.ve_layer_indices: + self.ve_shared = ValueEmbedding(vocab_size, ve_dim, kv_dim_ve) + self.ve_layer_scales = nn.ParameterList( + [nn.Parameter(torch.ones(1, dtype=torch.float32)) for _ in self.ve_layer_indices] + ) + else: + self.ve_shared = None + self.ve_layer_scales = nn.ParameterList() + self.value_embeds = nn.ModuleList() # keep empty for compat + self.final_norm = RMSNorm() + self.lm_head = None if tie_embeddings else CastedLinear(model_dim, vocab_size, bias=False) + if self.lm_head is not None: + self.lm_head._zero_init = True + self.mtp_heads = nn.ModuleList( + [CastedLinear(model_dim, vocab_size, bias=False) for _ in range(mtp_num_heads)] + ) + for head in self.mtp_heads: + head._zero_init = True + if xsa_last_n > 0: + for i in range(max(0, num_layers - xsa_last_n), num_layers): + self.blocks[i].attn.use_xsa = True + self._init_weights() + def _init_weights(self) -> None: + if self.tie_embeddings: + nn.init.normal_(self.tok_emb.weight, mean=0.0, std=self.tied_embed_init_std) + n = self.num_layers + proj_scale = 1.0 / math.sqrt(2 * n) + # Init banks: orthogonal, with proj layers scaled down and out/down zero-init + for i in range(n): + nn.init.orthogonal_(self.qo_bank.data[i], gain=1.0) # Q + nn.init.zeros_(self.qo_bank.data[n + i]) # Out (zero init) + nn.init.orthogonal_(self.kv_bank.data[i], gain=1.0) # K + nn.init.orthogonal_(self.kv_bank.data[n + i], gain=1.0) # V + nn.init.orthogonal_(self.mlp_up_bank.data[i], gain=1.0) # MLP up + nn.init.zeros_(self.mlp_down_bank.data[i]) # MLP down (zero init) + # Scale proj layers (out_proj and mlp_down are "proj" layers) + self.qo_bank.data[n + i].mul_(proj_scale) + self.mlp_down_bank.data[i].mul_(proj_scale) + # Init remaining nn.Linear modules (bigram proj, mtp heads, lm_head) + for name, module in self.named_modules(): + if isinstance(module, nn.Linear): + if getattr(module, "_zero_init", False): + nn.init.zeros_(module.weight) + elif module.weight.ndim == 2 and module.weight.shape[0] >= 64 and module.weight.shape[1] >= 64: + nn.init.orthogonal_(module.weight, gain=1.0) + def _get_ve(self, layer_idx: int, input_ids: Tensor, ve_cache: dict | None = None) -> Tensor | None: + """Get value embedding for a specific layer using shared table + per-layer scale.""" + if self.ve_shared is None or layer_idx not in self.ve_layer_indices: + return None + if ve_cache is not None and 've' not in ve_cache: + ve_cache['ve'] = self.ve_shared(input_ids) + ve_base = ve_cache['ve'] if ve_cache is not None else self.ve_shared(input_ids) + ve_idx = self.ve_layer_indices.index(layer_idx) + return ve_base * self.ve_layer_scales[ve_idx].to(dtype=ve_base.dtype) + def forward(self, input_ids: Tensor, target_ids: Tensor) -> Tensor: + n = self.num_layers + x = self.tok_emb(input_ids) + if self.bigram is not None: + x = x + self.bigram(input_ids) + x = F.rms_norm(x, (x.size(-1),)) + x = self.smear(x) + x0 = x + v0 = None + skips: list[Tensor] = [] + ve_cache: dict = {} + for i in range(self.num_encoder_layers): + ve = self._get_ve(i, input_ids, ve_cache) + x, raw_v = self.blocks[i](x, x0, + self.qo_bank[i], self.kv_bank[i], self.kv_bank[n + i], + self.qo_bank[n + i], self.mlp_up_bank[i], self.mlp_down_bank[i], + v_embed=ve, v0=v0) + if v0 is None and raw_v is not None: + v0 = raw_v + skips.append(x) + for i in range(self.num_decoder_layers): + bi = self.num_encoder_layers + i + if skips: + scaled_skip = self.skip_weights[i].to(dtype=x.dtype)[None, None, :] * skips.pop() + if self.skip_gates is not None: + g = torch.sigmoid(self.skip_gates[i].to(dtype=x.dtype))[None, None, :] + x = torch.lerp(scaled_skip, x, g) + else: + x = x + scaled_skip + ve = self._get_ve(bi, input_ids, ve_cache) + x, _ = self.blocks[bi](x, x0, + self.qo_bank[bi], self.kv_bank[bi], self.kv_bank[n + bi], + self.qo_bank[n + bi], self.mlp_up_bank[bi], self.mlp_down_bank[bi], + v_embed=ve, v0=v0) + x = self.final_norm(x) + x_flat = x.reshape(-1, x.size(-1)) + targets = target_ids.reshape(-1) + if self.tie_embeddings: + logits_proj = F.linear(x_flat, self.tok_emb.weight) + else: + if self.lm_head is None: + raise RuntimeError("lm_head is required when tie_embeddings=False") + logits_proj = self.lm_head(x_flat) + logits = self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + main_loss = F.cross_entropy(logits.float(), targets, reduction="mean") + if self.training and self.mtp_num_heads > 0 and self.mtp_loss_weight > 0.0: + _, seqlen, dim = x.shape + mtp_loss_sum = x.new_zeros(()) + mtp_loss_count = 0 + for k, mtp_head in enumerate(self.mtp_heads): + valid_t = seqlen - (k + 1) + if valid_t <= 0: + continue + mtp_hidden = x[:, :valid_t, :].reshape(-1, dim) + mtp_targets = target_ids[:, k + 1 :].reshape(-1) + mtp_logits_proj = mtp_head(mtp_hidden) + mtp_logits = self.logit_softcap * torch.tanh(mtp_logits_proj / self.logit_softcap) + mtp_loss_sum = mtp_loss_sum + F.cross_entropy(mtp_logits.float(), mtp_targets, reduction="mean") + mtp_loss_count += 1 + if mtp_loss_count > 0: + main_loss = main_loss + self.mtp_loss_weight * (mtp_loss_sum / mtp_loss_count) + return main_loss + def forward_hidden(self, input_ids: Tensor) -> Tensor: + """Return final hidden states before lm_head.""" + n = self.num_layers + x = self.tok_emb(input_ids) + if self.bigram is not None: + x = x + self.bigram(input_ids) + x = F.rms_norm(x, (x.size(-1),)) + x = self.smear(x) + x0 = x + v0 = None + skips: list[Tensor] = [] + ve_cache: dict = {} + for i in range(self.num_encoder_layers): + ve = self._get_ve(i, input_ids, ve_cache) + x, raw_v = self.blocks[i](x, x0, + self.qo_bank[i], self.kv_bank[i], self.kv_bank[n + i], + self.qo_bank[n + i], self.mlp_up_bank[i], self.mlp_down_bank[i], + v_embed=ve, v0=v0) + if v0 is None and raw_v is not None: + v0 = raw_v + skips.append(x) + for i in range(self.num_decoder_layers): + bi = self.num_encoder_layers + i + if skips: + scaled_skip = self.skip_weights[i].to(dtype=x.dtype)[None, None, :] * skips.pop() + if self.skip_gates is not None: + g = torch.sigmoid(self.skip_gates[i].to(dtype=x.dtype))[None, None, :] + x = torch.lerp(scaled_skip, x, g) + else: + x = x + scaled_skip + ve = self._get_ve(bi, input_ids, ve_cache) + x, _ = self.blocks[bi](x, x0, + self.qo_bank[bi], self.kv_bank[bi], self.kv_bank[n + bi], + self.qo_bank[n + bi], self.mlp_up_bank[bi], self.mlp_down_bank[bi], + v_embed=ve, v0=v0) + return self.final_norm(x) + + def compute_logits(self, hidden_states: Tensor) -> Tensor: + """Compute logits from hidden states.""" + if self.tie_embeddings: + logits_proj = F.linear(hidden_states, self.tok_emb.weight) + else: + logits_proj = self.lm_head(hidden_states) + return self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + + def forward_logits(self, input_ids: Tensor) -> Tensor: + """Return logits (bsz, seq_len, vocab) without computing loss.""" + return self.compute_logits(self.forward_hidden(input_ids)) + +# --- Sliding window evaluation --- + +def eval_val_sliding( + args: Hyperparameters, + base_model: nn.Module, + rank: int, + world_size: int, + device: torch.device, + val_tokens: Tensor, + base_bytes_lut: Tensor, + has_leading_space_lut: Tensor, + is_boundary_token_lut: Tensor, + stride: int, + batch_seqs: int = 32, + eval_seq_len: int | None = None, +) -> tuple[float, float]: + """Sliding window evaluation: each token scored with maximum context.""" + seq_len = eval_seq_len or args.train_seq_len + total_tokens = val_tokens.numel() - 1 + window_starts = [ws for ws in range(0, total_tokens, stride) + if min(ws + seq_len, total_tokens) - ws >= 1] + total_windows = len(window_starts) + my_s = (total_windows * rank) // world_size + my_e = (total_windows * (rank + 1)) // world_size + my_windows = window_starts[my_s:my_e] + loss_sum = torch.zeros((), device=device, dtype=torch.float64) + token_count = torch.zeros((), device=device, dtype=torch.float64) + byte_count = torch.zeros((), device=device, dtype=torch.float64) + base_model.eval() + compiled_logits = torch.compile(base_model.forward_logits, dynamic=False, fullgraph=True) + with torch.inference_mode(): + for bi in range(0, len(my_windows), batch_seqs): + batch_ws = my_windows[bi:bi + batch_seqs] + bsz = len(batch_ws) + x_batch = torch.zeros(bsz, seq_len, dtype=torch.int64, device=device) + y_batch = torch.zeros(bsz, seq_len, dtype=torch.int64, device=device) + wlens: list[int] = [] + for i, ws in enumerate(batch_ws): + end = min(ws + seq_len, total_tokens) + wlen = end - ws + wlens.append(wlen) + chunk = val_tokens[ws:end + 1].to(dtype=torch.int64, device=device) + x_batch[i, :wlen] = chunk[:-1] + y_batch[i, :wlen] = chunk[1:] + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + logits = compiled_logits(x_batch) + nll = F.cross_entropy( + logits.reshape(-1, logits.size(-1)).float(), + y_batch.reshape(-1), + reduction="none", + ).reshape(bsz, seq_len) + for i, ws in enumerate(batch_ws): + wlen = wlens[i] + s = 0 if ws == 0 else max(wlen - stride, 0) + scored_nll = nll[i, s:wlen].to(torch.float64) + loss_sum += scored_nll.sum() + token_count += float(wlen - s) + tgt = y_batch[i, s:wlen] + prev = x_batch[i, s:wlen] + tb = base_bytes_lut[tgt].to(torch.float64) + tb += (has_leading_space_lut[tgt] & ~is_boundary_token_lut[prev]).to(torch.float64) + byte_count += tb.sum() + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(token_count, op=dist.ReduceOp.SUM) + dist.all_reduce(byte_count, op=dist.ReduceOp.SUM) + val_loss = (loss_sum / token_count).item() + bits_per_token = val_loss / math.log(2.0) + tokens_per_byte = token_count.item() / byte_count.item() + base_model.train() + return val_loss, bits_per_token * tokens_per_byte + + +def ttt_adapt_adamw( + args, base_model: nn.Module, device: torch.device, + val_tokens: Tensor, rank: int = 0, world_size: int = 1, log0=print, +) -> None: + """AdamW TTT: fine-tune on val data BEFORE quantization (PR #1006 style).""" + seq_len = args.train_seq_len + total_seqs = (val_tokens.numel() - 1) // seq_len + batch_seqs = args.ttt_batch_seqs + if args.ttt_freeze_blocks > 0: + for i, block in enumerate(base_model.blocks): + if i < args.ttt_freeze_blocks: + for p in block.parameters(): + p.requires_grad_(False) + ttt_params = [p for p in base_model.parameters() if p.requires_grad] + log0(f"ttt_adamw:params trainable={sum(p.numel() for p in ttt_params)} " + f"frozen={sum(p.numel() for p in base_model.parameters() if not p.requires_grad)}") + optimizer = torch.optim.AdamW(ttt_params, lr=args.ttt_lr, weight_decay=0.0) + scheduler = None + if args.ttt_cosine_decay: + scheduler = torch.optim.lr_scheduler.CosineAnnealingLR( + optimizer, T_max=args.ttt_epochs, eta_min=args.ttt_lr * 0.1) + my_start = (total_seqs * rank) // world_size + my_end = (total_seqs * (rank + 1)) // world_size + base_model.train() + t0 = time.perf_counter() + for epoch in range(args.ttt_epochs): + epoch_loss_sum = torch.zeros((), device=device, dtype=torch.float64) + epoch_tokens = torch.zeros((), device=device, dtype=torch.float64) + for bs in range(my_start, my_end, batch_seqs): + be = min(bs + batch_seqs, my_end) + raw_start = bs * seq_len + raw_end = be * seq_len + 1 + if raw_end > val_tokens.numel(): + continue + local = val_tokens[raw_start:raw_end].to(device=device, dtype=torch.int64) + x = local[:-1].reshape(-1, seq_len) + y = local[1:].reshape(-1, seq_len) + optimizer.zero_grad(set_to_none=True) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + loss = base_model(x, y) + loss.backward() + if world_size > 1: + for p in ttt_params: + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG) + torch.nn.utils.clip_grad_norm_(ttt_params, args.ttt_grad_clip) + optimizer.step() + epoch_loss_sum += loss.detach().to(torch.float64) * float(y.numel()) + epoch_tokens += float(y.numel()) + if world_size > 1: + dist.all_reduce(epoch_loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(epoch_tokens, op=dist.ReduceOp.SUM) + epoch_avg_loss = epoch_loss_sum.item() / max(epoch_tokens.item(), 1) + if scheduler is not None: + scheduler.step() + log0(f"ttt_adamw:epoch {epoch+1}/{args.ttt_epochs} loss:{epoch_avg_loss:.4f} " + f"time:{time.perf_counter() - t0:.1f}s") + for p in base_model.parameters(): + p.requires_grad_(True) + base_model.eval() + log0(f"ttt_adamw:done elapsed={time.perf_counter() - t0:.1f}s") + + +def generate_autoregressive_calib(model, device, num_seqs=64, seq_len=2048, + vocab_size=1024, temperature=0.8, batch_size=8, seed=42): + """Generate sequences autoregressively from the model for GPTQ calibration. + No external data accessed — fully self-contained.""" + model.eval() + rng = torch.Generator(device=device) + rng.manual_seed(seed) + all_tokens = [] + with torch.inference_mode(), torch.autocast(device_type="cuda", dtype=torch.bfloat16): + for batch_start in range(0, num_seqs, batch_size): + bs = min(batch_size, num_seqs - batch_start) + tokens = torch.randint(0, vocab_size, (bs, 1), device=device, generator=rng) + for pos in range(seq_len - 1): + logits = model.forward_logits(tokens) + next_logit = logits[:, -1, :] + probs = torch.softmax(next_logit / temperature, dim=-1) + next_tok = torch.multinomial(probs, 1, generator=rng) + tokens = torch.cat([tokens, next_tok], dim=1) + for i in range(bs): + all_tokens.append(tokens[i:i+1]) + return all_tokens + + +def collect_hessians_from_tokens(hessian_model, token_seqs, device): + """Collect H = X^T X from pre-generated token sequences.""" + hessians = {} + hooks = [] + for name, module in hessian_model.named_modules(): + if isinstance(module, CastedLinear): + param_name = name + ".weight" + cols = module.weight.shape[1] + hessians[param_name] = torch.zeros(cols, cols, dtype=torch.float32, device='cpu') + h = module.register_forward_hook(_make_input_hook(hessians, param_name)) + hooks.append(h) + hessian_model.eval() + with torch.inference_mode(), torch.autocast(device_type="cuda", dtype=torch.bfloat16): + for seq in token_seqs: + x = seq[:, :-1].to(device) + y = seq[:, 1:].to(device) + hessian_model(x, y) + for h in hooks: + h.remove() + num_batches = len(token_seqs) + for name in hessians: + H = hessians[name] + H /= num_batches + damp = 0.01 * torch.diag(H).mean().clamp_min(1e-6) + H += damp * torch.eye(H.shape[0]) + hessians[name] = H + return hessians + + +# --- GPTQ-lite int6 quantization --- + +def _classify_param(name: str) -> str: + if "tok_emb" in name or "lm_head" in name: + return "embed" + if ".mlp." in name: + return "mlp" + if ".attn." in name or (".proj." in name and ".mlp." not in name): + return "attn" + return "other" +def sdclip_scale(t32: Tensor, clip_sigmas: float, clip_range: int) -> Tensor: + """SDClip: scale = clip_sigmas * std(row) / clip_range (per-row for 2D, global for 1D).""" + if t32.ndim == 2: + row_std = t32.std(dim=1).clamp_min(1e-10) + return (clip_sigmas * row_std / clip_range).clamp_min(1.0 / clip_range).to(torch.float16) + amax = t32.abs().max().item() + return torch.tensor(amax / clip_range if amax > 0 else 1.0, dtype=torch.float16) + +def quantize_int6_per_row(t: Tensor, clip_range: int = 31, clip_sigmas: float = 12.85) -> tuple[Tensor, Tensor]: + t32 = t.float() + s = sdclip_scale(t32, clip_sigmas, clip_range) + if t32.ndim == 2: + q = torch.clamp(torch.round(t32 / s.float()[:, None]), -clip_range, clip_range).to(torch.int8) + else: + q = torch.clamp(torch.round(t32 / s.float()), -clip_range, clip_range).to(torch.int8) + return q, s + +def quantize_int6_gptq(weight, hessian=None, clip_range=31, block_size=128, clip_sigmas=12.85): + """Full GPTQ: Hessian-aware int6 quantization with Cholesky error compensation. + Uses SDClip (k * std per row) for scale computation. + If hessian is None, falls back to SDClip-only quantization.""" + t32 = weight.float() + if t32.ndim != 2 or hessian is None: + return quantize_int6_per_row(t32, clip_range=clip_range, clip_sigmas=clip_sigmas) + rows, cols = t32.shape + H = hessian.float().clone() + dead = torch.diag(H) == 0 + H[dead, dead] = 1 + damp = 0.01 * torch.mean(torch.diag(H)) + H[torch.arange(cols), torch.arange(cols)] += damp + perm = torch.argsort(torch.diag(H), descending=True) + inv_perm = torch.argsort(perm) + W = t32[:, perm].clone() + W[:, dead[perm]] = 0 + H = H[perm][:, perm] + Hinv = torch.linalg.cholesky(H) + Hinv = torch.cholesky_inverse(Hinv) + Hinv = torch.linalg.cholesky(Hinv, upper=True) + # SDClip: single-pass scale from k * std(row) + s = sdclip_scale(t32, clip_sigmas, clip_range) + sf = s.float() + Q = torch.zeros_like(W, dtype=torch.int8) + W_work = W.clone() + for i1 in range(0, cols, block_size): + i2 = min(i1 + block_size, cols) + count = i2 - i1 + W1 = W_work[:, i1:i2].clone() + Q1 = torch.zeros(rows, count, dtype=torch.int8) + Err1 = torch.zeros(rows, count) + Hinv1 = Hinv[i1:i2, i1:i2] + for i in range(count): + w = W1[:, i] + d = Hinv1[i, i] + q = torch.clamp(torch.round(w / sf), -clip_range, clip_range).to(torch.int8) + Q1[:, i] = q + err = (w - q.float() * sf) / d + W1[:, i:] -= err.unsqueeze(1) * Hinv1[i, i:].unsqueeze(0) + Err1[:, i] = err + Q[:, i1:i2] = Q1 + if i2 < cols: + W_work[:, i2:] -= Err1 @ Hinv[i1:i2, i2:] + Q = Q[:, inv_perm] + return Q, s + +def _quantize_int6_percentile(t32, clip_range=31, clip_sigmas=12.85): + """Fallback: SDClip-based (kept for API compat, now uses SDClip not percentile).""" + return quantize_int6_per_row(t32, clip_range=clip_range, clip_sigmas=clip_sigmas) + +def _unbank_state_dict(sd: dict[str, Tensor], num_layers: int) -> dict[str, Tensor]: + """Convert 3D bank tensors into individual 2D tensors with standard names.""" + out: dict[str, Tensor] = {} + n = num_layers + for name, tensor in sd.items(): + if name == "qo_bank": + for i in range(n): + out[f"blocks.{i}.attn.c_q.weight"] = tensor[i] + out[f"blocks.{i}.attn.proj.weight"] = tensor[n + i] + elif name == "kv_bank": + for i in range(n): + out[f"blocks.{i}.attn.c_k.weight"] = tensor[i] + out[f"blocks.{i}.attn.c_v.weight"] = tensor[n + i] + elif name == "mlp_up_bank": + for i in range(n): + out[f"blocks.{i}.mlp.fc.weight"] = tensor[i] + elif name == "mlp_down_bank": + for i in range(n): + out[f"blocks.{i}.mlp.proj.weight"] = tensor[i] + else: + out[name] = tensor + return out + +def _rebank_state_dict(sd: dict[str, Tensor], num_layers: int, template_sd: dict[str, Tensor]) -> dict[str, Tensor]: + """Convert individual 2D tensors back into 3D bank tensors.""" + out: dict[str, Tensor] = {} + n = num_layers + # Reconstruct banks from individual weight keys + qo_slices = [None] * (2 * n) + kv_slices = [None] * (2 * n) + up_slices = [None] * n + down_slices = [None] * n + consumed = set() + for i in range(n): + qk = f"blocks.{i}.attn.c_q.weight" + if qk in sd: + qo_slices[i] = sd[qk] + consumed.add(qk) + ok = f"blocks.{i}.attn.proj.weight" + if ok in sd: + qo_slices[n + i] = sd[ok] + consumed.add(ok) + kk = f"blocks.{i}.attn.c_k.weight" + if kk in sd: + kv_slices[i] = sd[kk] + consumed.add(kk) + vk = f"blocks.{i}.attn.c_v.weight" + if vk in sd: + kv_slices[n + i] = sd[vk] + consumed.add(vk) + fk = f"blocks.{i}.mlp.fc.weight" + if fk in sd: + up_slices[i] = sd[fk] + consumed.add(fk) + dk = f"blocks.{i}.mlp.proj.weight" + if dk in sd: + down_slices[i] = sd[dk] + consumed.add(dk) + out["qo_bank"] = torch.stack(qo_slices).to(dtype=template_sd["qo_bank"].dtype) + out["kv_bank"] = torch.stack(kv_slices).to(dtype=template_sd["kv_bank"].dtype) + out["mlp_up_bank"] = torch.stack(up_slices).to(dtype=template_sd["mlp_up_bank"].dtype) + out["mlp_down_bank"] = torch.stack(down_slices).to(dtype=template_sd["mlp_down_bank"].dtype) + for name, tensor in sd.items(): + if name not in consumed: + out[name] = tensor + return out + +# --- Non-banked model for Hessian collection --- +# This mirrors the unbanked state dict keys: blocks.{i}.attn.c_q/c_k/c_v/proj, blocks.{i}.mlp.fc/proj + +class _HessianAttn(nn.Module): + """Non-banked attention with CastedLinear layers for Hessian hooks.""" + def __init__(self, dim, num_heads, num_kv_heads, rope_base, qk_gain_init): + super().__init__() + self.num_heads, self.num_kv_heads = num_heads, num_kv_heads + self.head_dim = dim // num_heads + kv_dim = num_kv_heads * self.head_dim + self.c_q = CastedLinear(dim, dim, bias=False) + self.c_k = CastedLinear(dim, kv_dim, bias=False) + self.c_v = CastedLinear(dim, kv_dim, bias=False) + self.proj = CastedLinear(dim, dim, bias=False) + self.q_gain = nn.Parameter(torch.full((num_heads,), qk_gain_init, dtype=torch.float32)) + self.rope_dims = 0 + self.rotary = Rotary(self.head_dim, base=rope_base, train_seq_len=1024) + self.use_xsa = False + def _xsa_efficient(self, y, v): + B, T, H, D = y.shape; Hkv = v.size(-2); group = H // Hkv + y_g = y.reshape(B, T, Hkv, group, D) + vn = F.normalize(v, dim=-1).unsqueeze(-2) + proj = (y_g * vn).sum(dim=-1, keepdim=True) * vn + return (y_g - proj).reshape(B, T, H, D) + def forward(self, x, v_embed=None): + bsz, seqlen, dim = x.shape + q = self.c_q(x).reshape(bsz, seqlen, self.num_heads, self.head_dim) + k = self.c_k(x).reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + v = self.c_v(x) + if v_embed is not None: + v = v + v_embed + v = v.reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + q = F.rms_norm(q, (q.size(-1),)) + k = F.rms_norm(k, (k.size(-1),)) + cos, sin = self.rotary(seqlen, x.device, q.dtype) + q = apply_rotary_emb(q, cos, sin, self.rope_dims) + k = apply_rotary_emb(k, cos, sin, self.rope_dims) + q = q * self.q_gain.to(dtype=q.dtype)[None, None, :, None] + if HAS_FA3: + y = flash_attn_3_func(q, k, v, causal=True) + else: + q2, k2, v2 = q.transpose(1, 2), k.transpose(1, 2), v.transpose(1, 2) + if k2.size(1) != q2.size(1): + reps = q2.size(1) // k2.size(1) + k2 = k2.repeat_interleave(reps, dim=1) + v2 = v2.repeat_interleave(reps, dim=1) + y = F.scaled_dot_product_attention(q2, k2, v2, is_causal=True).transpose(1, 2) + if self.use_xsa: + y = self._xsa_efficient(y, v) + return self.proj(y.reshape(bsz, seqlen, dim)) + +class _HessianMLP(nn.Module): + """Non-banked MLP with CastedLinear layers for Hessian hooks.""" + def __init__(self, dim, mlp_mult): + super().__init__() + self.fc = CastedLinear(dim, int(mlp_mult * dim), bias=False) + self.proj = CastedLinear(int(mlp_mult * dim), dim, bias=False) + def forward(self, x): + return self.proj(F.leaky_relu(self.fc(x), negative_slope=0.5).square()) + +class _HessianBlock(nn.Module): + def __init__(self, dim, num_heads, num_kv_heads, mlp_mult, rope_base, qk_gain_init, layer_idx=0, ln_scale=False): + super().__init__() + self.attn_norm = RMSNorm() + self.mlp_norm = RMSNorm() + self.attn = _HessianAttn(dim, num_heads, num_kv_heads, rope_base, qk_gain_init) + self.mlp = _HessianMLP(dim, mlp_mult) + self.attn_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.mlp_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.resid_mix = nn.Parameter(torch.stack((torch.ones(dim), torch.zeros(dim))).float()) + self.ln_scale_factor = 1.0 / math.sqrt(layer_idx + 1) if ln_scale else 1.0 + def forward(self, x, x0, v_embed=None): + mix = self.resid_mix.to(dtype=x.dtype) + x_in = mix[0][None, None, :] * x + mix[1][None, None, :] * x0 + attn_out = self.attn(self.attn_norm(x_in) * self.ln_scale_factor, v_embed=v_embed) + x_out = x_in + self.attn_scale.to(dtype=x_in.dtype)[None, None, :] * attn_out + x_out = x_out + self.mlp_scale.to(dtype=x_out.dtype)[None, None, :] * self.mlp(self.mlp_norm(x_out) * self.ln_scale_factor) + return x_out + +class _HessianGPT(nn.Module): + """Non-banked GPT model matching unbanked state dict keys for Hessian collection.""" + def __init__(self, vocab_size, num_layers, model_dim, num_heads, num_kv_heads, + mlp_mult, tie_embeddings, logit_softcap, rope_base, qk_gain_init, + bigram_vocab_size=0, bigram_dim=128, xsa_last_n=0, + rope_dims=0, ln_scale=False, + ve_enabled=False, ve_dim=128, ve_layers="9,10"): + super().__init__() + self.tie_embeddings = tie_embeddings + self.logit_softcap = logit_softcap + self.num_layers = num_layers + self.tok_emb = nn.Embedding(vocab_size, model_dim) + self.bigram = BigramHashEmbedding(bigram_vocab_size, bigram_dim, model_dim, trigram=bool(int(os.environ.get("TRIGRAM", "0")))) if bigram_vocab_size > 0 else None + self.smear = SmearGate(model_dim) + self.num_encoder_layers = num_layers // 2 + self.num_decoder_layers = num_layers - self.num_encoder_layers + self.num_skip_weights = min(self.num_encoder_layers, self.num_decoder_layers) + self.skip_weights = nn.Parameter(torch.ones(self.num_skip_weights, model_dim, dtype=torch.float32)) + self.blocks = nn.ModuleList([ + _HessianBlock(model_dim, num_heads, num_kv_heads, mlp_mult, rope_base, qk_gain_init, + layer_idx=i, ln_scale=ln_scale) + for i in range(num_layers) + ]) + if rope_dims > 0: + head_dim = model_dim // num_heads + for block in self.blocks: + block.attn.rope_dims = rope_dims + block.attn.rotary = Rotary(head_dim, base=rope_base, train_seq_len=1024, rope_dims=rope_dims) + if xsa_last_n > 0: + for i in range(max(0, num_layers - xsa_last_n), num_layers): + self.blocks[i].attn.use_xsa = True + kv_dim = num_kv_heads * (model_dim // num_heads) + self.ve_layer_indices = [int(x) for x in ve_layers.split(",") if x.strip()] if ve_enabled else [] + if self.ve_layer_indices: + self.ve_shared = ValueEmbedding(vocab_size, ve_dim, kv_dim) + self.ve_layer_scales = nn.ParameterList([nn.Parameter(torch.ones(1, dtype=torch.float32)) for _ in self.ve_layer_indices]) + else: + self.ve_shared = None + self.ve_layer_scales = nn.ParameterList() + self.final_norm = RMSNorm() + self.lm_head = None if tie_embeddings else CastedLinear(model_dim, vocab_size, bias=False) + def _get_ve(self, layer_idx, input_ids, ve_cache): + if self.ve_shared is None or layer_idx not in self.ve_layer_indices: + return None + if 've' not in ve_cache: + ve_cache['ve'] = self.ve_shared(input_ids) + ve_idx = self.ve_layer_indices.index(layer_idx) + return ve_cache['ve'] * self.ve_layer_scales[ve_idx].to(dtype=ve_cache['ve'].dtype) + def forward(self, input_ids, target_ids): + x = self.tok_emb(input_ids) + if self.bigram is not None: + x = x + self.bigram(input_ids) + x = F.rms_norm(x, (x.size(-1),)) + x = self.smear(x) + x0 = x + skips = [] + ve_cache = {} + for i in range(self.num_encoder_layers): + ve = self._get_ve(i, input_ids, ve_cache) + x = self.blocks[i](x, x0, v_embed=ve) + skips.append(x) + for i in range(self.num_decoder_layers): + bi = self.num_encoder_layers + i + if skips: + x = x + self.skip_weights[i].to(dtype=x.dtype)[None, None, :] * skips.pop() + ve = self._get_ve(bi, input_ids, ve_cache) + x = self.blocks[bi](x, x0, v_embed=ve) + x = self.final_norm(x) + x_flat = x.reshape(-1, x.size(-1)) + targets = target_ids.reshape(-1) + logits_proj = F.linear(x_flat, self.tok_emb.weight) if self.tie_embeddings else self.lm_head(x_flat) + logits = self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + return F.cross_entropy(logits.float(), targets, reduction="mean") + +def _make_input_hook(hessians, pname): + def hook_fn(module, input, output): + x = input[0].detach().float() + if x.ndim == 3: + x = x.reshape(-1, x.shape[-1]) + hessians[pname] += (x.T @ x).cpu() + return hook_fn + +def _make_output_hook(hessians, pname): + def hook_fn(module, input, output): + x = output.detach().float() + if x.ndim == 3: + x = x.reshape(-1, x.shape[-1]) + hessians[pname] += (x.T @ x).cpu() + return hook_fn + +def collect_hessians(hessian_model, train_loader, args, device, grad_accum_steps, num_batches=256): + """Run calibration batches through a non-banked model, collecting H = X^T X for each CastedLinear. + Also captures the pre-logit hidden states for tok_emb.weight GPTQ (embedding GPTQ).""" + hessians = {} + hooks = [] + for name, module in hessian_model.named_modules(): + if isinstance(module, CastedLinear): + param_name = name + ".weight" + cols = module.weight.shape[1] + hessians[param_name] = torch.zeros(cols, cols, dtype=torch.float32, device='cpu') + h = module.register_forward_hook(_make_input_hook(hessians, param_name)) + hooks.append(h) + # Embedding GPTQ: capture Hessian for tok_emb.weight via pre-logit hidden states. + # The tied embedding acts as a linear layer: logits = hidden @ tok_emb.weight.T + # so the activation Hessian is X^T X where X = final_norm output. + embed_hessian_key = "tok_emb.weight" + embed_dim = hessian_model.tok_emb.weight.shape[1] + hessians[embed_hessian_key] = torch.zeros(embed_dim, embed_dim, dtype=torch.float32, device='cpu') + embed_hook = hessian_model.final_norm.register_forward_hook( + _make_output_hook(hessians, embed_hessian_key) + ) + hooks.append(embed_hook) + hessian_model.eval() + with torch.inference_mode(), torch.autocast(device_type="cuda", dtype=torch.bfloat16): + for _ in range(num_batches): + x, y = train_loader.next_batch(args.train_batch_tokens, args.train_seq_len, grad_accum_steps) + hessian_model(x, y) + for h in hooks: + h.remove() + for name in hessians: + H = hessians[name] + H /= num_batches + damp = 0.01 * torch.diag(H).mean().clamp_min(1e-6) + H += damp * torch.eye(H.shape[0]) + hessians[name] = H + hessian_model.train() + return hessians + +def mixed_quantize_int6(state_dict: dict[str, Tensor], int6_cats: set[str], hessians: dict[str, Tensor] | None = None, + matrix_bits: int = 6, embed_bits: int = 8, + matrix_clip_sigmas: float = 12.85, embed_clip_sigmas: float = 20.0): + """Mixed-precision GPTQ quantization. + - mlp/attn weights: int6 with SDClip (k=matrix_clip_sigmas) and Hessian compensation + - tok_emb/lm_head: GPTQ with int8 and SDClip (k=embed_clip_sigmas) + - small tensors: passthrough float16 + """ + num_layers_total = max( + (int(k.split(".")[1]) for k in state_dict if k.startswith("blocks.")), + default=0, + ) + 1 + result: dict[str, Tensor] = {} + meta: dict[str, object] = {} + for name, tensor in state_dict.items(): + t = tensor.detach().cpu().contiguous() + cat = _classify_param(name) + if not t.is_floating_point() or t.numel() <= 65536: + result[name] = t.to(torch.float16) if t.is_floating_point() else t + meta[name] = "passthrough" + continue + if any(p in name for p in CONTROL_TENSOR_NAME_PATTERNS): + result[name] = t.float() + meta[name] = "passthrough_ctrl" + continue + if cat == "embed" and t.ndim >= 1: + # GPTQ on embeddings with int8 + SDClip + cr = 2 ** (embed_bits - 1) - 1 # 127 for int8 + H = hessians.get(name) if hessians else None + if H is not None: + q, s = quantize_int6_gptq(t, hessian=H, clip_range=cr, clip_sigmas=embed_clip_sigmas) + else: + q, s = quantize_int6_per_row(t, clip_range=cr, clip_sigmas=embed_clip_sigmas) + result[name + ".q"] = q + result[name + ".scale"] = s + meta[name] = {"type": f"int{embed_bits}"} + elif cat in int6_cats and t.ndim >= 1: + cr = 2 ** (matrix_bits - 1) - 1 # 31 for int6 + H = hessians.get(name) if hessians else None + if H is not None: + q, s = quantize_int6_gptq(t, hessian=H, clip_range=cr, clip_sigmas=matrix_clip_sigmas) + else: + q, s = quantize_int6_per_row(t, clip_range=cr, clip_sigmas=matrix_clip_sigmas) + result[name + ".q"] = q + result[name + ".scale"] = s + meta[name] = {"type": f"int{matrix_bits}"} + else: + q, s = quantize_float_tensor(t) + result[name + ".q"] = q + result[name + ".scale"] = s + meta[name] = {"type": "int8"} + return result, meta +def dequantize_mixed_int6(result: dict[str, Tensor], meta: dict[str, object], + template_sd: dict[str, Tensor]) -> dict[str, Tensor]: + out: dict[str, Tensor] = {} + for name, orig in template_sd.items(): + info = meta.get(name) + if info is None: + continue + orig_dtype = orig.dtype + if info in ("passthrough", "passthrough_ctrl", "passthrough_fp16"): + t = result[name] + if t.dtype == torch.float16 and orig_dtype in (torch.float32, torch.bfloat16): + t = t.to(orig_dtype) + out[name] = t + continue + q, s = result[name + ".q"], result[name + ".scale"] + if s.ndim > 0: + out[name] = (q.float() * s.float().view(q.shape[0], *([1] * (q.ndim - 1)))).to(orig_dtype) + else: + out[name] = (q.float() * float(s.item())).to(orig_dtype) + return out + +# --- Byte-shuffle + Brotli compression --- + +_BSHF_MAGIC = b'BSHF' + +def _byte_shuffle(data: bytes, stride: int = 2) -> bytes: + """Interleave bytes by stride before compression (improves Brotli ratio on quantized weights).""" + if stride <= 1 or len(data) < stride: + return data + src = np.frombuffer(data, dtype=np.uint8) + n = len(src) + out = np.empty(n, dtype=np.uint8) + dest_off = 0 + for pos in range(stride): + chunk = src[pos::stride] + out[dest_off:dest_off + len(chunk)] = chunk + dest_off += len(chunk) + return _BSHF_MAGIC + bytes([stride]) + out.tobytes() + +def _byte_unshuffle(data: bytes) -> bytes: + """Reverse byte shuffle.""" + if len(data) < 5 or data[:4] != _BSHF_MAGIC: + return data + stride = data[4] + if stride < 2: + return data[5:] + payload = np.frombuffer(data, dtype=np.uint8, offset=5) + n = len(payload) + out = np.empty(n, dtype=np.uint8) + src_off = 0 + for pos in range(stride): + chunk_len = n // stride + (1 if pos < n % stride else 0) + out[pos::stride][:chunk_len] = payload[src_off:src_off + chunk_len] + src_off += chunk_len + return out.tobytes() + +def _compress_brotli(data: bytes) -> bytes: + """Byte-shuffle then Brotli compress.""" + import brotli + shuffled = _byte_shuffle(data, stride=2) + return brotli.compress(shuffled, quality=11) + +def _decompress_brotli(data: bytes) -> bytes: + """Brotli decompress then byte-unshuffle.""" + import brotli + raw = brotli.decompress(data) + return _byte_unshuffle(raw) + +# --- Training --- + +def main() -> None: + code = Path(__file__).read_text(encoding="utf-8") + args = Hyperparameters() + # zeropower_via_newtonschulz5 runs eagerly with bmm -- do NOT compile + distributed = "RANK" in os.environ and "WORLD_SIZE" in os.environ + rank = int(os.environ.get("RANK", "0")) + world_size = int(os.environ.get("WORLD_SIZE", "1")) + local_rank = int(os.environ.get("LOCAL_RANK", "0")) + if world_size <= 0: + raise ValueError(f"WORLD_SIZE must be positive, got {world_size}") + if 8 % world_size != 0: + raise ValueError(f"WORLD_SIZE={world_size} must divide 8 so grad_accum_steps stays integral") + grad_accum_steps = 8 // world_size + grad_scale = 1.0 / grad_accum_steps + if not torch.cuda.is_available(): + raise RuntimeError("CUDA is required") + device = torch.device("cuda", local_rank) + torch.cuda.set_device(device) + if distributed: + dist.init_process_group(backend="nccl", device_id=device) + dist.barrier() + master_process = rank == 0 + torch.backends.cuda.matmul.allow_tf32 = True + torch.backends.cudnn.allow_tf32 = True + from torch.backends.cuda import enable_cudnn_sdp, enable_flash_sdp, enable_math_sdp, enable_mem_efficient_sdp + enable_cudnn_sdp(False) + enable_flash_sdp(True) + enable_mem_efficient_sdp(False) + enable_math_sdp(False) + logfile = None + if master_process: + os.makedirs("logs", exist_ok=True) + logfile = f"logs/{args.run_id}.txt" + print(logfile) + def log0(msg: str, console: bool = True) -> None: + if not master_process: + return + if console: + print(msg) + if logfile is not None: + with open(logfile, "a", encoding="utf-8") as f: + print(msg, file=f) + log0(code, console=False) + log0("=" * 100, console=False) + log0(f"Running Python {sys.version}", console=False) + log0(f"Running PyTorch {torch.__version__}", console=False) + log0( + subprocess.run(["nvidia-smi"], stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, check=False).stdout, + console=False, + ) + log0("=" * 100, console=False) + random.seed(args.seed) + np.random.seed(args.seed) + torch.manual_seed(args.seed) + torch.cuda.manual_seed_all(args.seed) + if not args.tokenizer_path.endswith(".model"): + raise ValueError(f"Script only setup for SentencePiece .model file: {args.tokenizer_path}") + sp = spm.SentencePieceProcessor(model_file=args.tokenizer_path) + if int(sp.vocab_size()) != args.vocab_size: + raise ValueError( + f"VOCAB_SIZE={args.vocab_size} does not match tokenizer vocab_size={int(sp.vocab_size())}" + ) + dataset_dir = Path(args.data_path).resolve() + actual_train_files = len(list(dataset_dir.glob("fineweb_train_*.bin"))) + effective_eval_seq_len = args.eval_seq_len if args.eval_seq_len > 0 else args.train_seq_len + val_seq_len = max(args.train_seq_len, effective_eval_seq_len) + val_tokens = load_validation_tokens(args.val_files, val_seq_len) + base_bytes_lut, has_leading_space_lut, is_boundary_token_lut = build_sentencepiece_luts( + sp, args.vocab_size, device + ) + log0(f"val_bpb:enabled tokenizer_kind=sentencepiece tokenizer_path={args.tokenizer_path}") + log0(f"train_loader:dataset:{dataset_dir.name} train_shards:{actual_train_files}") + log0(f"val_loader:shards pattern={args.val_files} tokens:{val_tokens.numel() - 1}") + CastedLinear._qat_enabled = args.qat_enabled + base_model = GPT( + vocab_size=args.vocab_size, + num_layers=args.num_layers, + model_dim=args.model_dim, + num_heads=args.num_heads, + num_kv_heads=args.num_kv_heads, + mlp_mult=args.mlp_mult, + tie_embeddings=args.tie_embeddings, + tied_embed_init_std=args.tied_embed_init_std, + logit_softcap=args.logit_softcap, + rope_base=args.rope_base, + qk_gain_init=args.qk_gain_init, + mtp_num_heads=args.mtp_num_heads, + mtp_loss_weight=args.mtp_loss_weight, + bigram_vocab_size=args.bigram_vocab_size, + bigram_dim=args.bigram_dim, + xsa_last_n=args.xsa_last_n, + rope_dims=args.rope_dims, + ln_scale=args.ln_scale, + dtg=args.dtg_enabled, + ve_enabled=args.ve_enabled, + ve_dim=args.ve_dim, + ve_layers=args.ve_layers, + gated_attention=args.gated_attention, + value_residual=args.value_residual, + skip_gates_enabled=args.skip_gates_enabled, + ).to(device).bfloat16() + # Banks stay FP32 (like CastedLinear weights), cast to BF16 in forward + base_model.qo_bank.data = base_model.qo_bank.data.float() + base_model.kv_bank.data = base_model.kv_bank.data.float() + base_model.mlp_up_bank.data = base_model.mlp_up_bank.data.float() + base_model.mlp_down_bank.data = base_model.mlp_down_bank.data.float() + for module in base_model.modules(): + if isinstance(module, CastedLinear): + module.float() + restore_low_dim_params_to_fp32(base_model) + # No DDP -- Parallel Muon handles bank grad communication via reduce-scatter, + # and non-bank grads are manually all-reduced before Adam steps. + compiled_model = torch.compile(base_model, dynamic=False, fullgraph=True) + model = compiled_model + + # Optimizer split: + # - 4 parameter banks -> Muon (batched Newton-Schulz) + # - token embedding -> Adam + # - scalars/control tensors -> Adam + # - bigram proj, mtp heads, VE proj -> Adam (small matrix params not worth banking) + matrix_params = [ + base_model.qo_bank, base_model.kv_bank, + base_model.mlp_up_bank, base_model.mlp_down_bank, + ] + block_named_params = list(base_model.blocks.named_parameters()) + scalar_params = [ + p + for name, p in block_named_params + if p.ndim < 2 or any(pattern in name for pattern in CONTROL_TENSOR_NAME_PATTERNS) + ] + if base_model.skip_weights.numel() > 0: + scalar_params.append(base_model.skip_weights) + if base_model.skip_gates is not None and base_model.skip_gates.numel() > 0: + scalar_params.append(base_model.skip_gates) + scalar_params.append(base_model.smear.gate) + if base_model.bigram is not None: + scalar_params.append(base_model.bigram.scale) + token_lr = args.tied_embed_lr if args.tie_embeddings else args.embed_lr + tok_params = [{"params": [base_model.tok_emb.weight], "lr": token_lr, "base_lr": token_lr}] + if base_model.bigram is not None: + tok_params.append({"params": [base_model.bigram.embed.weight], "lr": token_lr, "base_lr": token_lr}) + if base_model.bigram.proj is not None: + scalar_params.append(base_model.bigram.proj.weight) + if base_model.ve_shared is not None: + tok_params.append({"params": [base_model.ve_shared.embed.weight], "lr": token_lr, "base_lr": token_lr}) + if base_model.ve_shared.proj is not None: + scalar_params.append(base_model.ve_shared.proj.weight) + scalar_params.append(base_model.ve_shared.scale) + for s in base_model.ve_layer_scales: + scalar_params.append(s) + optimizer_tok = torch.optim.AdamW( + tok_params, + betas=(args.beta1, args.beta2), + eps=args.adam_eps, + weight_decay=args.embed_wd, + fused=True, + ) + optimizer_muon = Muon( + matrix_params, + lr=args.matrix_lr, + momentum=args.muon_momentum, + backend_steps=args.muon_backend_steps, + weight_decay=args.muon_wd, + ) + for group in optimizer_muon.param_groups: + group["base_lr"] = args.matrix_lr + optimizer_scalar = torch.optim.AdamW( + [{"params": scalar_params, "lr": args.scalar_lr, "base_lr": args.scalar_lr}], + betas=(args.beta1, args.beta2), + eps=args.adam_eps, + weight_decay=args.adam_wd, + fused=True, + ) + # Non-bank params that need manual all-reduce (replicated across GPUs) + replicated_params = list(optimizer_tok.param_groups[0]["params"]) + for pg in optimizer_tok.param_groups[1:]: + replicated_params.extend(pg["params"]) + replicated_params.extend(scalar_params) + + optimizer_head = None + if base_model.lm_head is not None: + optimizer_head = torch.optim.Adam( + [{"params": [base_model.lm_head.weight], "lr": args.head_lr, "base_lr": args.head_lr}], + betas=(args.beta1, args.beta2), + eps=args.adam_eps, + fused=True, + ) + replicated_params.append(base_model.lm_head.weight) + optimizers: list[torch.optim.Optimizer] = [optimizer_tok, optimizer_muon, optimizer_scalar] + if optimizer_head is not None: + optimizers.append(optimizer_head) + n_params = sum(p.numel() for p in base_model.parameters()) + mtp_params = sum(p.numel() for p in base_model.mtp_heads.parameters()) + log0(f"model_params:{n_params}") + log0(f"mtp_num_heads:{args.mtp_num_heads} mtp_loss_weight:{args.mtp_loss_weight} mtp_params:{mtp_params}") + xsa_layers = [i for i, b in enumerate(base_model.blocks) if b.attn.use_xsa] + log0(f"XSA:last_{args.xsa_last_n} active_layers:{xsa_layers}") + log0(f"world_size:{world_size} grad_accum_steps:{grad_accum_steps}") + log0("sdp_backends:cudnn=False flash=True mem_efficient=False math=False") + log0(f"attention_mode:gqa num_heads:{args.num_heads} num_kv_heads:{args.num_kv_heads}") + log0( + f"tie_embeddings:{args.tie_embeddings} embed_lr:{token_lr} " + f"head_lr:{args.head_lr if base_model.lm_head is not None else 0.0} " + f"matrix_lr:{args.matrix_lr} scalar_lr:{args.scalar_lr}" + ) + log0( + f"train_batch_tokens:{args.train_batch_tokens} train_seq_len:{args.train_seq_len} " + f"iterations:{args.iterations} warmup_steps:{args.warmup_steps} " + f"max_wallclock_seconds:{args.max_wallclock_seconds:.3f}" + ) + log0(f"seed:{args.seed}") + train_loader = DistributedTokenLoader(args.train_files, rank, world_size, device) + def zero_grad_all() -> None: + for opt in optimizers: + opt.zero_grad(set_to_none=True) + max_wallclock_ms = 1000.0 * args.max_wallclock_seconds if args.max_wallclock_seconds > 0 else None + def training_frac(step: int, elapsed_ms: float) -> float: + if max_wallclock_ms is None: + return step / max(args.iterations, 1) + return elapsed_ms / max(max_wallclock_ms, 1e-9) + def lr_mul(step: int, elapsed_ms: float) -> float: + # warmdown_frac > 0 takes priority; fallback to legacy warmdown_iters + if args.warmdown_frac > 0: + frac = training_frac(step, elapsed_ms) + if frac >= 1.0 - args.warmdown_frac: + return max((1.0 - frac) / args.warmdown_frac, 0.0) + return 1.0 + if args.warmdown_iters <= 0: + return 1.0 + if max_wallclock_ms is None: + warmdown_start = max(args.iterations - args.warmdown_iters, 0) + return max((args.iterations - step) / max(args.warmdown_iters, 1), 0.0) if warmdown_start <= step < args.iterations else 1.0 + step_ms = elapsed_ms / max(step, 1) + warmdown_ms = args.warmdown_iters * step_ms + remaining_ms = max(max_wallclock_ms - elapsed_ms, 0.0) + return remaining_ms / max(warmdown_ms, 1e-9) if remaining_ms <= warmdown_ms else 1.0 + if args.warmup_steps > 0: + initial_model_state = {name: tensor.detach().cpu().clone() for name, tensor in base_model.state_dict().items()} + initial_optimizer_states = [copy.deepcopy(opt.state_dict()) for opt in optimizers] + model.train() + for warmup_step in range(args.warmup_steps): + zero_grad_all() + for micro_step in range(grad_accum_steps): + x, y = train_loader.next_batch(args.train_batch_tokens, args.train_seq_len, grad_accum_steps) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + warmup_loss = model(x, y) + (warmup_loss * grad_scale).backward() + # All-reduce all grads for warmup (simple, not optimized) + if distributed: + for p in base_model.parameters(): + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG) + for opt in optimizers: + opt.step() + zero_grad_all() + if args.warmup_steps <= 20 or (warmup_step + 1) % 10 == 0 or warmup_step + 1 == args.warmup_steps: + log0(f"warmup_step:{warmup_step + 1}/{args.warmup_steps}") + base_model.load_state_dict(initial_model_state, strict=True) + for opt, state in zip(optimizers, initial_optimizer_states, strict=True): + opt.load_state_dict(state) + zero_grad_all() + train_loader = DistributedTokenLoader(args.train_files, rank, world_size, device) + swa_state: dict[str, Tensor] | None = None + swa_count = 0 + from collections import deque + lawa_queue: deque[dict[str, Tensor]] = deque(maxlen=args.lawa_k) + ema_state = {name: t.detach().float().clone() for name, t in base_model.state_dict().items()} + ema_decay = 0.997 + training_time_ms = 0.0 + stop_after_step: int | None = None + torch.cuda.synchronize() + t0 = time.perf_counter() + step = 0 + while True: + last_step = step == args.iterations or (stop_after_step is not None and step >= stop_after_step) + should_validate = last_step or (args.val_loss_every > 0 and step % args.val_loss_every == 0) + if should_validate: + torch.cuda.synchronize() + training_time_ms += 1000.0 * (time.perf_counter() - t0) + val_loss, val_bpb = eval_val( + args, + model, + rank, + world_size, + device, + grad_accum_steps, + val_tokens, + base_bytes_lut, + has_leading_space_lut, + is_boundary_token_lut, + ) + log0( + f"step:{step}/{args.iterations} val_loss:{val_loss:.4f} val_bpb:{val_bpb:.4f} " + f"train_time:{training_time_ms:.0f}ms step_avg:{training_time_ms / max(step, 1):.2f}ms" + ) + torch.cuda.synchronize() + t0 = time.perf_counter() + if last_step: + if stop_after_step is not None and step < args.iterations: + log0( + f"stopping_early: wallclock_cap train_time:{training_time_ms:.0f}ms " + f"step:{step}/{args.iterations}" + ) + break + elapsed_ms = training_time_ms + 1000.0 * (time.perf_counter() - t0) + scale = lr_mul(step, elapsed_ms) + if args.late_qat_threshold > 0 and scale < args.late_qat_threshold and not CastedLinear._qat_enabled: + CastedLinear._qat_enabled = True + log0(f"late_qat:enabled step:{step} scale:{scale:.4f}") + zero_grad_all() + train_loss = torch.zeros((), device=device) + for micro_step in range(grad_accum_steps): + x, y = train_loader.next_batch(args.train_batch_tokens, args.train_seq_len, grad_accum_steps) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + loss = model(x, y) + train_loss += loss.detach() + (loss * grad_scale).backward() + train_loss /= grad_accum_steps + frac = min(step / args.muon_momentum_warmup_steps, 1.0) if args.muon_momentum_warmup_steps > 0 else 1.0 + muon_momentum = (1 - frac) * args.muon_momentum_warmup_start + frac * args.muon_momentum + for group in optimizer_muon.param_groups: + group["momentum"] = muon_momentum + for opt in optimizers: + for group in opt.param_groups: + group["lr"] = group["base_lr"] * scale + if args.grad_clip_norm > 0: + torch.nn.utils.clip_grad_norm_(base_model.parameters(), args.grad_clip_norm) + # === 3-phase overlapped optimizer step === + # Phase 1: Launch async reduce-scatter for banks (biggest first) + optimizer_muon.launch_reduce_scatters() + # Phase 2: All-reduce non-bank grads + step Adam (while bank RS is in-flight) + if distributed: + for p in replicated_params: + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG) + optimizer_tok.step() + optimizer_scalar.step() + if optimizer_head is not None: + optimizer_head.step() + # Phase 3: Wait for RS, local NS5, all-gather (banks processed last) + optimizer_muon.step() + zero_grad_all() + # EMA update + with torch.no_grad(): + for name, t in base_model.state_dict().items(): + ema_state[name].mul_(ema_decay).add_(t.detach().float(), alpha=1.0 - ema_decay) + step += 1 + approx_training_time_ms = training_time_ms + 1000.0 * (time.perf_counter() - t0) + if args.swa_enabled and scale < 0.2 and step % args.swa_every == 0: + if swa_state is None: + swa_state = {name: t.detach().cpu().clone() for name, t in base_model.state_dict().items()} + swa_count = 1 + log0(f"swa:start step:{step}") + else: + for name, t in base_model.state_dict().items(): + swa_state[name] += t.detach().cpu() + swa_count += 1 + if args.lawa_enabled and step % args.lawa_freq == 0: + lawa_queue.append({name: t.detach().cpu().clone() for name, t in base_model.state_dict().items()}) + should_log_train = ( + args.train_log_every > 0 + and (step <= 10 or step % args.train_log_every == 0 or stop_after_step is not None) + ) + if should_log_train: + log0( + f"step:{step}/{args.iterations} train_loss:{train_loss.item():.4f} " + f"train_time:{approx_training_time_ms:.0f}ms step_avg:{approx_training_time_ms / step:.2f}ms" + ) + reached_cap = max_wallclock_ms is not None and approx_training_time_ms >= max_wallclock_ms + if distributed and max_wallclock_ms is not None: + reached_cap_tensor = torch.tensor(int(reached_cap), device=device) + dist.all_reduce(reached_cap_tensor, op=dist.ReduceOp.MAX) + reached_cap = bool(reached_cap_tensor.item()) + if stop_after_step is None and reached_cap: + stop_after_step = step + log0( + f"peak memory allocated: {torch.cuda.max_memory_allocated() // 1024 // 1024} MiB " + f"reserved: {torch.cuda.max_memory_reserved() // 1024 // 1024} MiB" + ) + # Apply weight averaging + if args.lawa_enabled and len(lawa_queue) > 1: + log0(f"lawa:applying LAWA averaging k={len(lawa_queue)}") + current_state = base_model.state_dict() + avg_state = {name: torch.zeros(t.shape, dtype=torch.float32, device='cpu') for name, t in current_state.items()} + for snap in lawa_queue: + for name in avg_state: + avg_state[name] += snap[name].float() + for name in avg_state: + avg_state[name] /= len(lawa_queue) + avg_state[name] = avg_state[name].to(dtype=current_state[name].dtype) + base_model.load_state_dict(avg_state, strict=True) + else: + log0("ema:applying EMA weights") + current_state = base_model.state_dict() + avg_state = {name: t.to(dtype=current_state[name].dtype) for name, t in ema_state.items()} + base_model.load_state_dict(avg_state, strict=True) + torch.cuda.synchronize() + t_diag = time.perf_counter() + diag_val_loss, diag_val_bpb = eval_val( + args, compiled_model, rank, world_size, device, grad_accum_steps, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + ) + torch.cuda.synchronize() + log0( + f"DIAGNOSTIC post_ema val_loss:{diag_val_loss:.4f} val_bpb:{diag_val_bpb:.4f} " + f"eval_time:{1000.0 * (time.perf_counter() - t_diag):.0f}ms" + ) + # AdamW TTT: fine-tune EMA model on val data BEFORE quantization + if args.ttt_enabled: + if distributed: + dist.barrier() + log0(f"ttt:start lr={args.ttt_lr} epochs={args.ttt_epochs} " + f"freeze_blocks={args.ttt_freeze_blocks} cosine_decay={args.ttt_cosine_decay}") + t_ttt = time.perf_counter() + ttt_adapt_adamw( + args, base_model, device, val_tokens, + rank=rank, world_size=world_size, log0=log0, + ) + torch.cuda.synchronize() + log0(f"ttt:elapsed={time.perf_counter() - t_ttt:.1f}s") + t_ttt_diag = time.perf_counter() + ttt_diag_loss, ttt_diag_bpb = eval_val( + args, base_model, rank, world_size, device, grad_accum_steps, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + ) + torch.cuda.synchronize() + log0(f"DIAGNOSTIC post_ttt val_loss:{ttt_diag_loss:.4f} val_bpb:{ttt_diag_bpb:.4f} " + f"eval_time:{1000.0 * (time.perf_counter() - t_ttt_diag):.0f}ms") + if distributed: + dist.barrier() + full_state_dict = base_model.state_dict() + export_sd = {k: v for k, v in full_state_dict.items() if "mtp_heads" not in k} + excluded_mtp = sum(int(t.numel()) for k, t in full_state_dict.items() if "mtp_heads" in k) + if excluded_mtp > 0: + log0(f"export_excluding_mtp_params:{excluded_mtp}") + if master_process: + torch.save(export_sd, "final_model.pt") + model_bytes = os.path.getsize("final_model.pt") + code_bytes = len(code.encode("utf-8")) + log0(f"Serialized model: {model_bytes} bytes") + log0(f"Code size: {code_bytes} bytes") + # Unbank 3D tensors into individual 2D tensors for quantization + sd_cpu = {k: v.detach().cpu() for k, v in export_sd.items()} + unbanked_sd = _unbank_state_dict(sd_cpu, args.num_layers) + # Full GPTQ: collect Hessians via a temporary non-banked model + log0(f"gptq:building non-banked model for Hessian collection...") + hessian_model = _HessianGPT( + vocab_size=args.vocab_size, num_layers=args.num_layers, model_dim=args.model_dim, + num_heads=args.num_heads, num_kv_heads=args.num_kv_heads, mlp_mult=args.mlp_mult, + tie_embeddings=args.tie_embeddings, logit_softcap=args.logit_softcap, + rope_base=args.rope_base, qk_gain_init=args.qk_gain_init, + bigram_vocab_size=args.bigram_vocab_size, bigram_dim=args.bigram_dim, + xsa_last_n=args.xsa_last_n, rope_dims=args.rope_dims, ln_scale=args.ln_scale, + ve_enabled=args.ve_enabled, ve_dim=args.ve_dim, ve_layers=args.ve_layers, + ).to(device).bfloat16() + for m in hessian_model.modules(): + if isinstance(m, CastedLinear): + m.float() + restore_low_dim_params_to_fp32(hessian_model) + # Load unbanked weights into the non-banked model + hessian_model.load_state_dict( + {k: v.to(device) for k, v in unbanked_sd.items() if k in hessian_model.state_dict()}, + strict=False, + ) + # Training-data calibration (proven -0.0007 BPP vs AR self-gen) + log0(f"gptq:collecting hessians from training data ({args.gptq_calib_batches} batches)...") + hessians = collect_hessians(hessian_model, train_loader, args, device, grad_accum_steps, num_batches=args.gptq_calib_batches) + log0(f"gptq:collected hessians for {len(hessians)} layers (training data)") + del hessian_model + torch.cuda.empty_cache() + quant_result, quant_meta = mixed_quantize_int6( + unbanked_sd, {"mlp", "attn"}, hessians=hessians, + matrix_bits=args.matrix_bits, embed_bits=args.embed_bits, + matrix_clip_sigmas=args.matrix_clip_sigmas, embed_clip_sigmas=args.embed_clip_sigmas, + ) + # NOVEL: Selective ±1 pruning by reconstruction error + # Sort ±1 quantized values by their reconstruction error (scale²), + # prune least-impactful first until artifact fits target size. + target_mb = float(os.environ.get("TARGET_MB", "15.9")) + code_bytes_est = len(code.encode("utf-8")) + ones_info = [] # (tensor_key, flat_idx, error) + for name, info in quant_meta.items(): + if not (isinstance(info, dict) and info.get("type") == "int6"): continue + qk, sk = name + ".q", name + ".scale" + if qk not in quant_result or sk not in quant_result: continue + q, s = quant_result[qk], quant_result[sk] + if s.ndim > 0: + ones_mask = (q.abs() == 1) + if ones_mask.any(): + row_idx = torch.arange(q.shape[0]).unsqueeze(1).expand_as(q)[ones_mask] + flat_idx = torch.arange(q.numel()).reshape(q.shape)[ones_mask] + errors = s.float()[row_idx].pow(2) + for fi, err in zip(flat_idx.tolist(), errors.tolist()): + ones_info.append((qk, fi, err)) + if ones_info: + ones_info.sort(key=lambda x: x[2]) + def _try_prune(n): + tmp = {k: v.clone() for k, v in quant_result.items()} + for i in range(min(n, len(ones_info))): + tmp[ones_info[i][0]].view(-1)[ones_info[i][1]] = 0 + buf = io.BytesIO(); torch.save({"w": tmp, "m": quant_meta}, buf) + return len(_compress_brotli(buf.getvalue())) + code_bytes_est, tmp + no_sz, _ = _try_prune(0) + target_bytes = int(target_mb * 1024 * 1024) + log0(f"selective_prune: {len(ones_info)} ±1 candidates, unpruned={no_sz/(1024*1024):.2f}MB target={target_mb}MB") + if no_sz <= target_bytes: + log0("selective_prune: already fits, no pruning needed") + else: + full_sz, _ = _try_prune(len(ones_info)) + log0(f"selective_prune: full ±1 prune={full_sz/(1024*1024):.2f}MB") + if full_sz > target_bytes: + log0("selective_prune: even full prune not enough, applying all") + _, quant_result = _try_prune(len(ones_info)) + else: + lo, hi = 0, len(ones_info) + while lo < hi: + mid = (lo + hi) // 2 + sz, _ = _try_prune(mid) + if sz <= target_bytes: hi = mid + else: lo = mid + 1 + log0(f"selective_prune: pruning {lo}/{len(ones_info)} ±1 values ({100*lo/len(ones_info):.1f}%) to fit {target_mb}MB") + _, quant_result = _try_prune(lo) + quant_buf = io.BytesIO() + torch.save({"w": quant_result, "m": quant_meta}, quant_buf) + quant_raw = quant_buf.getvalue() + quant_blob = _compress_brotli(quant_raw) + if master_process: + with open("final_model.int6.ptz", "wb") as f: + f.write(quant_blob) + quant_file_bytes = len(quant_blob) + code_bytes = len(code.encode("utf-8")) + log0(f"Serialized model int6+brotli+byteshuffle: {quant_file_bytes} bytes") + log0(f"Total submission size int6+brotli+byteshuffle: {quant_file_bytes + code_bytes} bytes") + if distributed: + dist.barrier() + with open("final_model.int6.ptz", "rb") as f: + quant_blob_disk = f.read() + quant_state = torch.load( + io.BytesIO(_decompress_brotli(quant_blob_disk)), + map_location="cpu", + weights_only=False, + ) + deq_unbanked = dequantize_mixed_int6(quant_state["w"], quant_state["m"], unbanked_sd) + # Re-bank the dequantized tensors + deq_state = _rebank_state_dict(deq_unbanked, args.num_layers, sd_cpu) + eval_model = GPT( + vocab_size=args.vocab_size, num_layers=args.num_layers, model_dim=args.model_dim, + num_heads=args.num_heads, num_kv_heads=args.num_kv_heads, mlp_mult=args.mlp_mult, + tie_embeddings=args.tie_embeddings, tied_embed_init_std=args.tied_embed_init_std, + logit_softcap=args.logit_softcap, rope_base=args.rope_base, qk_gain_init=args.qk_gain_init, + mtp_num_heads=0, mtp_loss_weight=0.0, + bigram_vocab_size=args.bigram_vocab_size, bigram_dim=args.bigram_dim, + xsa_last_n=args.xsa_last_n, + rope_dims=args.rope_dims, ln_scale=args.ln_scale, dtg=args.dtg_enabled, + ve_enabled=args.ve_enabled, ve_dim=args.ve_dim, ve_layers=args.ve_layers, + gated_attention=args.gated_attention, value_residual=args.value_residual, + skip_gates_enabled=args.skip_gates_enabled, + ).to(device).bfloat16() + eval_model.qo_bank.data = eval_model.qo_bank.data.float() + eval_model.kv_bank.data = eval_model.kv_bank.data.float() + eval_model.mlp_up_bank.data = eval_model.mlp_up_bank.data.float() + eval_model.mlp_down_bank.data = eval_model.mlp_down_bank.data.float() + for m in eval_model.modules(): + if isinstance(m, CastedLinear): + m.float() + restore_low_dim_params_to_fp32(eval_model) + eval_model.load_state_dict(deq_state, strict=True) + compiled_eval = torch.compile(eval_model, dynamic=False, fullgraph=True) + torch.cuda.synchronize() + t_qeval = time.perf_counter() + q_val_loss, q_val_bpb = eval_val( + args, compiled_eval, rank, world_size, device, grad_accum_steps, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + eval_seq_len=effective_eval_seq_len, + ) + torch.cuda.synchronize() + log0( + f"final_int6_roundtrip val_loss:{q_val_loss:.4f} val_bpb:{q_val_bpb:.4f} " + f"eval_time:{1000.0 * (time.perf_counter() - t_qeval):.0f}ms" + ) + log0(f"final_int6_roundtrip_exact val_loss:{q_val_loss:.8f} val_bpb:{q_val_bpb:.8f}") + sw_seq_len = effective_eval_seq_len + if args.eval_stride > 0 and args.eval_stride < sw_seq_len: + torch.cuda.synchronize() + t_slide = time.perf_counter() + sw_val_loss, sw_val_bpb = eval_val_sliding( + args, eval_model, rank, world_size, device, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + stride=args.eval_stride, + eval_seq_len=sw_seq_len, + ) + torch.cuda.synchronize() + log0( + f"final_int6_sliding_window val_loss:{sw_val_loss:.4f} val_bpb:{sw_val_bpb:.4f} " + f"stride:{args.eval_stride} eval_time:{1000.0 * (time.perf_counter() - t_slide):.0f}ms" + ) + log0(f"final_int6_sliding_window_exact val_loss:{sw_val_loss:.8f} val_bpb:{sw_val_bpb:.8f}") + log0(f"final_int8_zlib_roundtrip_exact val_loss:{sw_val_loss:.8f} val_bpb:{sw_val_bpb:.8f}") + if args.eval_stride != 64 and 64 < sw_seq_len: + torch.cuda.synchronize() + t_slide64 = time.perf_counter() + sw64_val_loss, sw64_val_bpb = eval_val_sliding( + args, eval_model, rank, world_size, device, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + stride=64, + eval_seq_len=sw_seq_len, + ) + torch.cuda.synchronize() + log0( + f"final_int6_sliding_window_s64 val_loss:{sw64_val_loss:.4f} val_bpb:{sw64_val_bpb:.4f} " + f"stride:64 eval_time:{1000.0 * (time.perf_counter() - t_slide64):.0f}ms" + ) + log0(f"final_int6_sliding_window_s64_exact val_loss:{sw64_val_loss:.8f} val_bpb:{sw64_val_bpb:.8f}") + log0(f"final_int8_zlib_roundtrip_exact val_loss:{sw64_val_loss:.8f} val_bpb:{sw64_val_bpb:.8f}") + # --- Causal SLOT: per-batch delta optimization using ONLY context (already-scored) positions --- + if args.slot_enabled and args.eval_stride > 0 and args.eval_stride < sw_seq_len: + try: + slot_stride = args.eval_stride + seq_s = sw_seq_len + total_tok = val_tokens.numel() - 1 + ws_list = [ws for ws in range(0, total_tok, slot_stride) if min(ws + seq_s, total_tok) - ws >= 1] + my_s = (len(ws_list) * rank) // world_size + my_e = (len(ws_list) * (rank + 1)) // world_size + my_ws = ws_list[my_s:my_e] + num_batches = (len(my_ws) + 31) // 32 + sl_loss = torch.zeros((), device=device, dtype=torch.float64) + sl_tc = torch.zeros((), device=device, dtype=torch.float64) + sl_bc = torch.zeros((), device=device, dtype=torch.float64) + torch.cuda.synchronize() + t_slot = time.perf_counter() + eval_model.eval() + log0(f"causal_slot:start lr={args.slot_lr} steps={args.slot_steps} stride={slot_stride} " + f"windows={len(my_ws)} batches={num_batches}") + for batch_idx, bi in enumerate(range(0, len(my_ws), 32)): + bws = my_ws[bi:bi+32]; bsz = len(bws) + xb = torch.zeros(bsz, seq_s, dtype=torch.int64, device=device) + yb = torch.zeros(bsz, seq_s, dtype=torch.int64, device=device) + wls = [] + for i, ws in enumerate(bws): + end = min(ws + seq_s, total_tok); wl = end - ws; wls.append(wl) + ct = val_tokens[ws:end+1].to(dtype=torch.int64, device=device) + xb[i,:wl] = ct[:-1]; yb[i,:wl] = ct[1:] + # Frozen forward: get hidden states + with torch.no_grad(), torch.autocast(device_type="cuda", dtype=torch.bfloat16): + H = eval_model.forward_hidden(xb) + H = H.detach().float() + # Build context mask (already-scored positions) and score mask (new positions) + context_mask = torch.zeros(bsz, seq_s, dtype=torch.bool, device=device) + has_context = False + for i, ws in enumerate(bws): + wl = wls[i]; s = 0 if ws == 0 else max(wl - slot_stride, 0) + if s > 0: + context_mask[i, :s] = True + has_context = True + # Optimize delta on context positions only (CAUSAL: no future info) + delta = torch.zeros(1, 1, H.shape[-1], device=device, dtype=H.dtype, requires_grad=True) + if has_context: + sopt = torch.optim.AdamW([delta], lr=args.slot_lr, betas=(0.3, 0.9), weight_decay=1e-8, eps=1e-5) + for _ in range(args.slot_steps): + sopt.zero_grad() + lg = eval_model.compute_logits((H + delta).to(torch.bfloat16)).float() + nll_all = F.cross_entropy(lg.reshape(-1, lg.size(-1)), yb.reshape(-1), reduction="none").reshape(bsz, seq_s) + ctx_nll = nll_all[context_mask] + if ctx_nll.numel() > 0: + loss_c = ctx_nll.mean() + loss_c.backward() + sopt.step() + # Score new positions with adapted delta + with torch.no_grad(): + lg = eval_model.compute_logits((H + delta.detach()).to(torch.bfloat16)).float() + nll = F.cross_entropy(lg.reshape(-1, lg.size(-1)), yb.reshape(-1), reduction="none").reshape(bsz, seq_s) + for i, ws in enumerate(bws): + wl = wls[i]; s = 0 if ws == 0 else max(wl - slot_stride, 0) + sl_loss += nll[i, s:wl].to(torch.float64).sum(); sl_tc += float(wl - s) + tgt, prev = yb[i, s:wl], xb[i, s:wl] + tb = base_bytes_lut[tgt].to(torch.float64) + tb += (has_leading_space_lut[tgt] & ~is_boundary_token_lut[prev]).to(torch.float64) + sl_bc += tb.sum() + if batch_idx % 500 == 0 or batch_idx == num_batches - 1: + log0(f" causal_slot:batch {batch_idx+1}/{num_batches} " + f"time:{time.perf_counter()-t_slot:.1f}s"); sys.stdout.flush() + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(sl_loss, op=dist.ReduceOp.SUM) + dist.all_reduce(sl_tc, op=dist.ReduceOp.SUM) + dist.all_reduce(sl_bc, op=dist.ReduceOp.SUM) + sv_loss = (sl_loss / sl_tc).item() + sv_bpb = sv_loss / math.log(2.0) * (sl_tc.item() / sl_bc.item()) + torch.cuda.synchronize() + log0(f"final_causal_slot val_loss:{sv_loss:.4f} val_bpb:{sv_bpb:.4f} time:{1000*(time.perf_counter()-t_slot):.0f}ms") + log0(f"final_causal_slot_exact val_loss:{sv_loss:.8f} val_bpb:{sv_bpb:.8f}") + log0(f"final_int8_zlib_roundtrip_exact val_loss:{sv_loss:.8f} val_bpb:{sv_bpb:.8f}") + except Exception as e: + import traceback; log0(f"causal_slot:ERROR {e}"); traceback.print_exc(); sys.stdout.flush() + if distributed: + dist.destroy_process_group() + sys.exit(0) +if __name__ == "__main__": + main() + +==================================================================================================== +Running Python 3.12.3 (main, Nov 6 2025, 13:44:16) [GCC 13.3.0] +Running PyTorch 2.9.1+cu128 +Mon Apr 6 05:22:25 2026 ++-----------------------------------------------------------------------------------------+ +| NVIDIA-SMI 580.126.09 Driver Version: 580.126.09 CUDA Version: 13.0 | ++-----------------------------------------+------------------------+----------------------+ +| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC | +| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. | +| | | MIG M. | +|=========================================+========================+======================| +| 0 NVIDIA H100 80GB HBM3 On | 00000000:19:00.0 Off | 0 | +| N/A 41C P0 121W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 1 NVIDIA H100 80GB HBM3 On | 00000000:3B:00.0 Off | 0 | +| N/A 33C P0 117W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 2 NVIDIA H100 80GB HBM3 On | 00000000:4C:00.0 Off | 0 | +| N/A 32C P0 117W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 3 NVIDIA H100 80GB HBM3 On | 00000000:5D:00.0 Off | 0 | +| N/A 40C P0 123W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 4 NVIDIA H100 80GB HBM3 On | 00000000:9B:00.0 Off | 0 | +| N/A 41C P0 121W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 5 NVIDIA H100 80GB HBM3 On | 00000000:BB:00.0 Off | 0 | +| N/A 35C P0 118W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 6 NVIDIA H100 80GB HBM3 On | 00000000:CB:00.0 Off | 0 | +| N/A 40C P0 121W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ +| 7 NVIDIA H100 80GB HBM3 On | 00000000:DB:00.0 Off | 0 | +| N/A 32C P0 111W / 700W | 1521MiB / 81559MiB | 0% Default | +| | | Disabled | ++-----------------------------------------+------------------------+----------------------+ + ++-----------------------------------------------------------------------------------------+ +| Processes: | +| GPU GI CI PID Type Process name GPU Memory | +| ID ID Usage | +|=========================================================================================| +| No running processes found | ++-----------------------------------------------------------------------------------------+ + +==================================================================================================== +val_bpb:enabled tokenizer_kind=sentencepiece tokenizer_path=./data/tokenizers/fineweb_8192_bpe.model +train_loader:dataset:fineweb10B_sp8192 train_shards:128 +val_loader:shards pattern=./data/datasets/fineweb10B_sp8192/fineweb_val_*.bin tokens:40540160 +model_params:35941976 +mtp_num_heads:0 mtp_loss_weight:0.2 mtp_params:0 +XSA:last_11 active_layers:[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10] +world_size:8 grad_accum_steps:1 +sdp_backends:cudnn=False flash=True mem_efficient=False math=False +attention_mode:gqa num_heads:8 num_kv_heads:4 +tie_embeddings:True embed_lr:0.035 head_lr:0.0 matrix_lr:0.025 scalar_lr:0.025 +train_batch_tokens:786432 train_seq_len:2048 iterations:20000 warmup_steps:20 max_wallclock_seconds:600.000 +seed:42 +warmup_step:1/20 +warmup_step:2/20 +warmup_step:3/20 +warmup_step:4/20 +warmup_step:5/20 +warmup_step:6/20 +warmup_step:7/20 +warmup_step:8/20 +warmup_step:9/20 +warmup_step:10/20 +warmup_step:11/20 +warmup_step:12/20 +warmup_step:13/20 +warmup_step:14/20 +warmup_step:15/20 +warmup_step:16/20 +warmup_step:17/20 +warmup_step:18/20 +warmup_step:19/20 +warmup_step:20/20 +step:0/20000 val_loss:9.0007 val_bpb:3.4845 train_time:0ms step_avg:0.01ms +step:1/20000 train_loss:9.0016 train_time:145ms step_avg:145.31ms +step:2/20000 train_loss:9.8631 train_time:173ms step_avg:86.43ms +step:3/20000 train_loss:9.9185 train_time:269ms step_avg:89.59ms +step:4/20000 train_loss:9.3452 train_time:367ms step_avg:91.69ms +step:5/20000 train_loss:8.7939 train_time:464ms step_avg:92.86ms +step:6/20000 train_loss:8.6242 train_time:562ms step_avg:93.62ms +step:7/20000 train_loss:8.3552 train_time:664ms step_avg:94.81ms +step:8/20000 train_loss:7.8692 train_time:763ms step_avg:95.32ms +step:9/20000 train_loss:7.8115 train_time:862ms step_avg:95.82ms +step:10/20000 train_loss:7.5881 train_time:963ms step_avg:96.28ms +step:500/20000 train_loss:3.3427 train_time:49909ms step_avg:99.82ms +step:1000/20000 train_loss:3.0749 train_time:100085ms step_avg:100.08ms +step:1500/20000 train_loss:3.2815 train_time:150197ms step_avg:100.13ms +step:2000/20000 train_loss:3.1159 train_time:200257ms step_avg:100.13ms +step:2500/20000 train_loss:3.2618 train_time:250317ms step_avg:100.13ms +step:3000/20000 train_loss:2.9688 train_time:300325ms step_avg:100.11ms +step:3500/20000 train_loss:3.1116 train_time:350308ms step_avg:100.09ms +step:4000/20000 train_loss:3.0288 train_time:400243ms step_avg:100.06ms +step:4000/20000 val_loss:3.0164 val_bpb:1.1677 train_time:400309ms step_avg:100.08ms +step:4500/20000 train_loss:3.0471 train_time:450267ms step_avg:100.06ms +step:5000/20000 train_loss:2.8813 train_time:500189ms step_avg:100.04ms +swa:start step:5200 +late_qat:enabled step:5395 scale:0.1498 +step:5500/20000 train_loss:2.8768 train_time:550638ms step_avg:100.12ms +step:5987/20000 val_loss:2.8509 val_bpb:1.1037 train_time:600121ms step_avg:100.24ms +stopping_early: wallclock_cap train_time:600121ms step:5987/20000 +peak memory allocated: 26621 MiB reserved: 26678 MiB +ema:applying EMA weights +DIAGNOSTIC post_ema val_loss:2.8473 val_bpb:1.1023 eval_time:1596ms +ttt:start lr=0.0005 epochs=6 freeze_blocks=2 cosine_decay=True +ttt_adamw:params trainable=35937864 frozen=4112 +ttt_adamw:epoch 1/6 loss:2.9090 time:14.2s +ttt_adamw:epoch 2/6 loss:2.8349 time:28.2s +ttt_adamw:epoch 3/6 loss:2.8120 time:42.2s +ttt_adamw:epoch 4/6 loss:2.7929 time:56.2s +ttt_adamw:epoch 5/6 loss:2.7769 time:70.2s +ttt_adamw:epoch 6/6 loss:2.7658 time:84.2s +ttt_adamw:done elapsed=84.2s +ttt:elapsed=84.2s +DIAGNOSTIC post_ttt val_loss:2.7595 val_bpb:1.0683 eval_time:4728ms +Serialized model: 135395695 bytes +Code size: 116724 bytes +gptq:building non-banked model for Hessian collection... +gptq:collecting hessians from training data (256 batches)... +gptq:collected hessians for 67 layers (training data) +selective_prune: 9821142 ±1 candidates, unpruned=14.42MB target=15.9MB +selective_prune: already fits, no pruning needed +Serialized model int6+brotli+byteshuffle: 14998505 bytes +Total submission size int6+brotli+byteshuffle: 15115229 bytes +final_int6_roundtrip val_loss:2.8248 val_bpb:1.0936 eval_time:6136ms +final_int6_roundtrip_exact val_loss:2.82482546 val_bpb:1.09357797 +final_int6_sliding_window val_loss:2.7879 val_bpb:1.0793 stride:64 eval_time:67263ms +final_int6_sliding_window_exact val_loss:2.78786651 val_bpb:1.07927455 +final_int8_zlib_roundtrip_exact val_loss:2.78786651 val_bpb:1.07927455 diff --git a/train_gpt.py b/train_gpt.py index 651beb2b89..21804aa28f 100644 --- a/train_gpt.py +++ b/train_gpt.py @@ -1,14 +1,8 @@ -""" -The `train_gpt.py` and `train_gpt_mlx.py` scripts are intended as good launching-off points for new participants, not SOTA configs. We'll accept PRs that tune, improve, or simplify these scripts without significantly increasing complexity, but competitive submissions should stay in the `/records` folder. - -Hard stop: To keep readable for newcomers, let's make sure `train_gpt.py` and `train_gpt_mlx.py` never are longer than 1500 lines. -""" - from __future__ import annotations - import copy import glob import io +import lzma import math import os import random @@ -18,7 +12,11 @@ import uuid import zlib from pathlib import Path - +try: + import zstandard + _COMPRESSOR = "zstd" +except ImportError: + _COMPRESSOR = "zlib" import numpy as np import sentencepiece as spm import torch @@ -26,156 +24,268 @@ import torch.nn.functional as F from torch import Tensor, nn from torch.nn.parallel import DistributedDataParallel as DDP - -# ----------------------------- -# HYPERPARAMETERS -# ----------------------------- -# Default Simple Baseline run: -# - 9 transformer blocks at width 512 -# - 8 attention heads with 4 KV heads (GQA) and 2x MLP expansion -# - vocab size 1024, sequence length 1024, tied embeddings -# - 524,288 train tokens per step for 20,000 iterations with a ~10 minute cap - +try: + from flash_attn_interface import flash_attn_func as flash_attn_3_func + HAS_FA3 = True +except ImportError: + HAS_FA3 = False class Hyperparameters: - # Data paths are shard globs produced by the existing preprocessing pipeline. - data_path = os.environ.get("DATA_PATH", "./data/datasets/fineweb10B_sp1024") + data_path = os.environ.get("DATA_PATH", "./data/datasets/fineweb10B_sp8192") train_files = os.path.join(data_path, "fineweb_train_*.bin") val_files = os.path.join(data_path, "fineweb_val_*.bin") - tokenizer_path = os.environ.get("TOKENIZER_PATH", "./data/tokenizers/fineweb_1024_bpe.model") + tokenizer_path = os.environ.get("TOKENIZER_PATH", "./data/tokenizers/fineweb_8192_bpe.model") run_id = os.environ.get("RUN_ID", str(uuid.uuid4())) seed = int(os.environ.get("SEED", 1337)) - - # Validation cadence and batch size. Validation always uses the full fineweb_val split. val_batch_size = int(os.environ.get("VAL_BATCH_SIZE", 524_288)) - val_loss_every = int(os.environ.get("VAL_LOSS_EVERY", 1000)) - train_log_every = int(os.environ.get("TRAIN_LOG_EVERY", 200)) - - # Training length. + val_loss_every = int(os.environ.get("VAL_LOSS_EVERY", 4000)) + train_log_every = int(os.environ.get("TRAIN_LOG_EVERY", 500)) iterations = int(os.environ.get("ITERATIONS", 20000)) - warmdown_iters = int(os.environ.get("WARMDOWN_ITERS", 1200)) + warmdown_iters = int(os.environ.get("WARMDOWN_ITERS", 3500)) # legacy; overridden by warmdown_frac if > 0 + warmdown_frac = float(os.environ.get("WARMDOWN_FRAC", 0.667)) # fraction of total time for warmdown warmup_steps = int(os.environ.get("WARMUP_STEPS", 20)) - train_batch_tokens = int(os.environ.get("TRAIN_BATCH_TOKENS", 524_288)) - train_seq_len = int(os.environ.get("TRAIN_SEQ_LEN", 1024)) + train_batch_tokens = int(os.environ.get("TRAIN_BATCH_TOKENS", 786_432)) + train_seq_len = int(os.environ.get("TRAIN_SEQ_LEN", 2048)) + eval_seq_len = int(os.environ.get("EVAL_SEQ_LEN", 2048)) max_wallclock_seconds = float(os.environ.get("MAX_WALLCLOCK_SECONDS", 600.0)) - qk_gain_init = float(os.environ.get("QK_GAIN_INIT", 1.5)) - - # Model shape. - vocab_size = int(os.environ.get("VOCAB_SIZE", 1024)) - num_layers = int(os.environ.get("NUM_LAYERS", 9)) + qk_gain_init = float(os.environ.get("QK_GAIN_INIT", 4.0)) + vocab_size = int(os.environ.get("VOCAB_SIZE", 8192)) + num_layers = int(os.environ.get("NUM_LAYERS", 11)) num_kv_heads = int(os.environ.get("NUM_KV_HEADS", 4)) model_dim = int(os.environ.get("MODEL_DIM", 512)) num_heads = int(os.environ.get("NUM_HEADS", 8)) - mlp_mult = int(os.environ.get("MLP_MULT", 2)) + mlp_mult = float(os.environ.get("MLP_MULT", 4.0)) tie_embeddings = bool(int(os.environ.get("TIE_EMBEDDINGS", "1"))) rope_base = float(os.environ.get("ROPE_BASE", 10000.0)) logit_softcap = float(os.environ.get("LOGIT_SOFTCAP", 30.0)) - - # Optimizer hyperparameters. embed_lr = float(os.environ.get("EMBED_LR", 0.6)) head_lr = float(os.environ.get("HEAD_LR", 0.008)) - tied_embed_lr = float(os.environ.get("TIED_EMBED_LR", 0.05)) + tied_embed_lr = float(os.environ.get("TIED_EMBED_LR", 0.035)) tied_embed_init_std = float(os.environ.get("TIED_EMBED_INIT_STD", 0.005)) - matrix_lr = float(os.environ.get("MATRIX_LR", 0.04)) - scalar_lr = float(os.environ.get("SCALAR_LR", 0.04)) - muon_momentum = float(os.environ.get("MUON_MOMENTUM", 0.95)) + matrix_lr = float(os.environ.get("MATRIX_LR", 0.025)) + scalar_lr = float(os.environ.get("SCALAR_LR", 0.025)) + muon_momentum = float(os.environ.get("MUON_MOMENTUM", 0.99)) muon_backend_steps = int(os.environ.get("MUON_BACKEND_STEPS", 5)) - muon_momentum_warmup_start = float(os.environ.get("MUON_MOMENTUM_WARMUP_START", 0.85)) - muon_momentum_warmup_steps = int(os.environ.get("MUON_MOMENTUM_WARMUP_STEPS", 500)) + muon_momentum_warmup_start = float(os.environ.get("MUON_MOMENTUM_WARMUP_START", 0.92)) + muon_momentum_warmup_steps = int(os.environ.get("MUON_MOMENTUM_WARMUP_STEPS", 1500)) beta1 = float(os.environ.get("BETA1", 0.9)) beta2 = float(os.environ.get("BETA2", 0.95)) adam_eps = float(os.environ.get("ADAM_EPS", 1e-8)) - grad_clip_norm = float(os.environ.get("GRAD_CLIP_NORM", 0.0)) - -# ----------------------------- -# MUON OPTIMIZER -# ----------------------------- -# -# As borrowed from modded-nanogpt -# Background on Muon: https://kellerjordan.github.io/posts/muon/ - -def zeropower_via_newtonschulz5(G: Tensor, steps: int = 10, eps: float = 1e-7) -> Tensor: - # Orthogonalize a 2D update matrix with a fast Newton-Schulz iteration. - # Muon uses this to normalize matrix-shaped gradients before applying them. + grad_clip_norm = float(os.environ.get("GRAD_CLIP_NORM", 0.3)) + eval_stride = int(os.environ.get("EVAL_STRIDE", 64)) + mtp_num_heads = int(os.environ.get("MTP_NUM_HEADS", 0)) + mtp_loss_weight = float(os.environ.get("MTP_LOSS_WEIGHT", 0.2)) + muon_beta2 = float(os.environ.get("MUON_BETA2", 0.95)) + swa_enabled = bool(int(os.environ.get("SWA_ENABLED", "1"))) + swa_every = int(os.environ.get("SWA_EVERY", 50)) + lawa_enabled = bool(int(os.environ.get("LAWA_ENABLED", "0"))) + lawa_k = int(os.environ.get("LAWA_K", 10)) + lawa_freq = int(os.environ.get("LAWA_FREQ", 100)) + muon_wd = float(os.environ.get("MUON_WD", 0.085)) + adam_wd = float(os.environ.get("ADAM_WD", 0.02)) + embed_wd = float(os.environ.get("EMBED_WD", 0.085)) + qat_enabled = bool(int(os.environ.get("QAT_ENABLED", "0"))) + bigram_vocab_size = int(os.environ.get("BIGRAM_VOCAB_SIZE", 0)) # disabled: SP8192 replaces BigramHash + bigram_dim = int(os.environ.get("BIGRAM_DIM", 128)) + trigram_enabled = bool(int(os.environ.get("TRIGRAM", "0"))) # TrigramHash (off by default, risky) + xsa_last_n = int(os.environ.get("XSA_LAST_N", 11)) # XSA on ALL layers (our novel contribution) + rope_dims = int(os.environ.get("ROPE_DIMS", 16)) + ln_scale = bool(int(os.environ.get("LN_SCALE", "1"))) + dtg_enabled = bool(int(os.environ.get("DTG_ENABLED", "0"))) + late_qat_threshold = float(os.environ.get("LATE_QAT_THRESHOLD", 0.15)) + ve_enabled = bool(int(os.environ.get("VE_ENABLED", "0"))) + ve_dim = int(os.environ.get("VE_DIM", 128)) + ve_layers = os.environ.get("VE_LAYERS", "9,10") + gated_attention = bool(int(os.environ.get("GATED_ATTENTION", "0"))) + value_residual = bool(int(os.environ.get("VALUE_RESIDUAL", "0"))) # VRL with sigmoid gates (off by default, risky) + skip_gates_enabled = bool(int(os.environ.get("SKIP_GATES_ENABLED", "1"))) # learned sigmoid gates on U-Net skips + # GPTQ calibration + gptq_calib_batches = int(os.environ.get("GPTQ_CALIB_BATCHES", 256)) + gptq_block_size = int(os.environ.get("GPTQ_BLOCK_SIZE", 128)) + # SDClip: clip threshold = k * std(row) instead of percentile search + matrix_bits = int(os.environ.get("MATRIX_BITS", 6)) + embed_bits = int(os.environ.get("EMBED_BITS", 8)) + matrix_clip_sigmas = float(os.environ.get("MATRIX_CLIP_SIGMAS", 12.85)) # k for int6 matrices + embed_clip_sigmas = float(os.environ.get("EMBED_CLIP_SIGMAS", 20.0)) # k for int8 embeddings + # TTT: AdamW test-time training (pre-quantization, on EMA model) + ttt_enabled = bool(int(os.environ.get("TTT_ENABLED", "1"))) + ttt_lr = float(os.environ.get("TTT_LR", 0.0005)) + ttt_epochs = int(os.environ.get("TTT_EPOCHS", 6)) + ttt_batch_seqs = int(os.environ.get("TTT_BATCH_SEQS", 32)) + ttt_freeze_blocks = int(os.environ.get("TTT_FREEZE_BLOCKS", 2)) + ttt_grad_clip = float(os.environ.get("TTT_GRAD_CLIP", 1.0)) + ttt_cosine_decay = bool(int(os.environ.get("TTT_COSINE_DECAY", "1"))) + # Causal SLOT: per-batch delta optimization at last hidden layer (post-quantization) + slot_enabled = bool(int(os.environ.get("SLOT_ENABLED", "0"))) + slot_lr = float(os.environ.get("SLOT_LR", 0.005)) + slot_steps = int(os.environ.get("SLOT_STEPS", 8)) + +# --- Batched Newton-Schulz orthogonalization --- + +def zeropower_via_newtonschulz5(G: Tensor, steps: int = 5, eps: float = 1e-7) -> Tensor: + """Batched Newton-Schulz orthogonalization. G: (B,M,N) or (M,N).""" a, b, c = (3.4445, -4.7750, 2.0315) + was_2d = G.ndim == 2 + if was_2d: + G = G.unsqueeze(0) X = G.bfloat16() - X /= X.norm() + eps - transposed = G.size(0) > G.size(1) + transposed = X.size(-2) > X.size(-1) if transposed: - X = X.T + X = X.mT + X = X / (X.norm(dim=(-2, -1), keepdim=True) + eps) for _ in range(steps): - A = X @ X.T - B = b * A + c * A @ A + A = X @ X.mT + B = b * A + c * (A @ A) X = a * X + B @ X - return X.T if transposed else X + if transposed: + X = X.mT + if was_2d: + X = X.squeeze(0) + return X +# --- Parallel Muon optimizer --- class Muon(torch.optim.Optimizer): - def __init__(self, params, lr: float, momentum: float, backend_steps: int, nesterov: bool = True): + """Parallel Muon: post-backward reduce-scatter -> local NS5 -> all-gather. + + No DDP for bank params. After backward, this optimizer: + 1. Launches async reduce-scatter for all banks (biggest first) + 2. Returns control so Adam can step on small params while RS is in-flight + 3. Waits for each RS, runs local NS5 on the shard, launches async all-gather + 4. Each all-gather overlaps with next bank's NS5 + """ + def __init__(self, params, lr: float, momentum: float, backend_steps: int, + nesterov: bool = True, weight_decay: float = 0.0): super().__init__( params, - dict(lr=lr, momentum=momentum, backend_steps=backend_steps, nesterov=nesterov), + dict(lr=lr, momentum=momentum, backend_steps=backend_steps, + nesterov=nesterov, weight_decay=weight_decay), ) + self._built = False + + def _build(self): + self._distributed = dist.is_available() and dist.is_initialized() + self._world_size = dist.get_world_size() if self._distributed else 1 + self._rank = dist.get_rank() if self._distributed else 0 + ws = self._world_size + + self._bank_meta = [] + for group in self.param_groups: + for p in group["params"]: + B = p.shape[0] + padded_B = ((B + ws - 1) // ws) * ws + shard_B = padded_B // ws + tail = p.shape[1:] + dev = p.device + self._bank_meta.append({ + 'p': p, + 'B': B, + 'padded_grad': torch.zeros(padded_B, *tail, device=dev, dtype=torch.bfloat16), + 'shard': torch.zeros(shard_B, *tail, device=dev, dtype=torch.bfloat16), + 'shard_mom': torch.zeros(shard_B, *tail, device=dev, dtype=torch.bfloat16), + 'full_update': torch.zeros(padded_B, *tail, device=dev, dtype=torch.bfloat16), + 'scale': max(1, p.shape[-2] / p.shape[-1]) ** 0.5, + }) + # Sort by size descending -- launch biggest reduce-scatters first + self._bank_meta.sort(key=lambda m: -m['p'].numel()) + self._built = True + + def launch_reduce_scatters(self): + """Phase 1: launch async reduce-scatter for all banks. Call right after backward.""" + if not self._built: + self._build() + if not self._distributed: + return + self._rs_futures = [] + for m in self._bank_meta: + p = m['p'] + if p.grad is None: + self._rs_futures.append(None) + continue + pg = m['padded_grad'] + pg[:m['B']].copy_(p.grad.bfloat16()) + if pg.shape[0] > m['B']: + pg[m['B']:].zero_() + fut = dist.reduce_scatter_tensor(m['shard'], pg, op=dist.ReduceOp.AVG, async_op=True) + self._rs_futures.append(fut) @torch.no_grad() def step(self, closure=None): + """Phase 3: wait for RS, local NS5, all-gather. Call AFTER Adam steps.""" loss = None if closure is not None: with torch.enable_grad(): loss = closure() - distributed = dist.is_available() and dist.is_initialized() - world_size = dist.get_world_size() if distributed else 1 - rank = dist.get_rank() if distributed else 0 + if not self._built: + self._build() for group in self.param_groups: - params = group["params"] - if not params: - continue lr = group["lr"] momentum = group["momentum"] backend_steps = group["backend_steps"] nesterov = group["nesterov"] - - total_params = sum(int(p.numel()) for p in params) - updates_flat = torch.zeros(total_params, device=params[0].device, dtype=torch.bfloat16) - - curr = 0 - for i, p in enumerate(params): - if i % world_size == rank and p.grad is not None: - g = p.grad + wd = group.get("weight_decay", 0.0) + + prev_ag_handle = None + prev_m = None + + sharded = self._distributed and hasattr(self, '_rs_futures') + + for i, m in enumerate(self._bank_meta): + p = m['p'] + if p.grad is None: + continue + + if prev_ag_handle is not None: + prev_ag_handle.wait() + pp = prev_m['p'] + upd = prev_m['full_update'][:prev_m['B']] + if wd > 0.0: + pp.data.mul_(1.0 - lr * wd) + pp.add_(upd.to(dtype=pp.dtype), alpha=-lr * prev_m['scale']) + + if sharded and self._rs_futures[i] is not None: + self._rs_futures[i].wait() + g = m['shard'] + buf = m['shard_mom'] + else: + g = p.grad.bfloat16() state = self.state[p] if "momentum_buffer" not in state: state["momentum_buffer"] = torch.zeros_like(g) buf = state["momentum_buffer"] - buf.mul_(momentum).add_(g) - if nesterov: - g = g.add(buf, alpha=momentum) - g = zeropower_via_newtonschulz5(g, steps=backend_steps) - # Scale correction from Muon reference implementations. - g *= max(1, g.size(0) / g.size(1)) ** 0.5 - updates_flat[curr : curr + p.numel()] = g.reshape(-1) - curr += p.numel() - - if distributed: - dist.all_reduce(updates_flat, op=dist.ReduceOp.SUM) - curr = 0 - for p in params: - g = updates_flat[curr : curr + p.numel()].view_as(p).to(dtype=p.dtype) - p.add_(g, alpha=-lr) - curr += p.numel() + buf.mul_(momentum).add_(g) + if nesterov: + update = g.add(buf, alpha=momentum) + else: + update = buf + + update = zeropower_via_newtonschulz5(update, steps=backend_steps) + + if sharded: + prev_ag_handle = dist.all_gather_into_tensor( + m['full_update'], update, async_op=True) + prev_m = m + else: + if wd > 0.0: + p.data.mul_(1.0 - lr * wd) + p.add_(update.to(dtype=p.dtype), alpha=-lr * m['scale']) + + if prev_ag_handle is not None: + prev_ag_handle.wait() + pp = prev_m['p'] + upd = prev_m['full_update'][:prev_m['B']] + if wd > 0.0: + pp.data.mul_(1.0 - lr * wd) + pp.add_(upd.to(dtype=pp.dtype), alpha=-lr * prev_m['scale']) + + if hasattr(self, '_rs_futures'): + del self._rs_futures return loss - -# ----------------------------- -# TOKENIZER-AGNOSTIC EVALUATION SETUP -# ----------------------------- -# -# It's common for small models have a large fraction of their parameters be embeddings, since the 2 * d_model * d_vocab vectors can be gigantic. -# Instead of locking the tokenizer, we let you bring your own and calculate our validation metrics on the average compression of the validation set. -# We calculate BPB (bits-per-byte) instead of validation loss, so we need methods to count the number of bits per token in the tokenizer. -# Note: Submissions that edit the tokenizer will be examined more carefully, since screwing this up might unjustly improve your score. +# --- Tokenizer evaluation helpers --- def build_sentencepiece_luts( sp: spm.SentencePieceProcessor, vocab_size: int, device: torch.device @@ -193,7 +303,7 @@ def build_sentencepiece_luts( base_bytes_np[token_id] = 1 continue piece = sp.id_to_piece(token_id) - if piece.startswith("▁"): + if piece.startswith("\u2581"): has_leading_space_np[token_id] = True piece = piece[1:] base_bytes_np[token_id] = len(piece.encode("utf-8")) @@ -202,20 +312,15 @@ def build_sentencepiece_luts( torch.tensor(has_leading_space_np, dtype=torch.bool, device=device), torch.tensor(is_boundary_token_np, dtype=torch.bool, device=device), ) - - def load_validation_tokens(pattern: str, seq_len: int) -> Tensor: files = [Path(p) for p in sorted(glob.glob(pattern))] if not files: raise FileNotFoundError(f"No files found for pattern: {pattern}") - # The export pipeline writes the fixed first-50k-doc validation set to fineweb_val_*. tokens = torch.cat([load_data_shard(file) for file in files]).contiguous() usable = ((tokens.numel() - 1) // seq_len) * seq_len if usable <= 0: raise ValueError(f"Validation split is too short for TRAIN_SEQ_LEN={seq_len}") return tokens[: usable + 1] - - def eval_val( args: Hyperparameters, model: nn.Module, @@ -227,34 +332,32 @@ def eval_val( base_bytes_lut: Tensor, has_leading_space_lut: Tensor, is_boundary_token_lut: Tensor, + eval_seq_len: int | None = None, ) -> tuple[float, float]: - # Validation computes two metrics: - # - val_loss: token cross-entropy (natural log) - # - val_bpb: tokenizer-agnostic compression metric used by the challenge + seq_len = eval_seq_len or args.train_seq_len local_batch_tokens = args.val_batch_size // (world_size * grad_accum_steps) - if local_batch_tokens < args.train_seq_len: + if local_batch_tokens < seq_len: raise ValueError( "VAL_BATCH_SIZE must provide at least one sequence per rank; " f"got VAL_BATCH_SIZE={args.val_batch_size}, WORLD_SIZE={world_size}, " - f"GRAD_ACCUM_STEPS={grad_accum_steps}, TRAIN_SEQ_LEN={args.train_seq_len}" + f"GRAD_ACCUM_STEPS={grad_accum_steps}, seq_len={seq_len}" ) - local_batch_seqs = local_batch_tokens // args.train_seq_len - total_seqs = (val_tokens.numel() - 1) // args.train_seq_len + local_batch_seqs = local_batch_tokens // seq_len + total_seqs = (val_tokens.numel() - 1) // seq_len seq_start = (total_seqs * rank) // world_size seq_end = (total_seqs * (rank + 1)) // world_size val_loss_sum = torch.zeros((), device=device, dtype=torch.float64) val_token_count = torch.zeros((), device=device, dtype=torch.float64) val_byte_count = torch.zeros((), device=device, dtype=torch.float64) - model.eval() with torch.inference_mode(): for batch_seq_start in range(seq_start, seq_end, local_batch_seqs): batch_seq_end = min(batch_seq_start + local_batch_seqs, seq_end) - raw_start = batch_seq_start * args.train_seq_len - raw_end = batch_seq_end * args.train_seq_len + 1 + raw_start = batch_seq_start * seq_len + raw_end = batch_seq_end * seq_len + 1 local = val_tokens[raw_start:raw_end].to(device=device, dtype=torch.int64, non_blocking=True) - x = local[:-1].reshape(-1, args.train_seq_len) - y = local[1:].reshape(-1, args.train_seq_len) + x = local[:-1].reshape(-1, seq_len) + y = local[1:].reshape(-1, seq_len) with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): batch_loss = model(x, y).detach() batch_token_count = float(y.numel()) @@ -265,31 +368,23 @@ def eval_val( token_bytes = base_bytes_lut[tgt_ids].to(dtype=torch.int16) token_bytes += (has_leading_space_lut[tgt_ids] & ~is_boundary_token_lut[prev_ids]).to(dtype=torch.int16) val_byte_count += token_bytes.to(torch.float64).sum() - if dist.is_available() and dist.is_initialized(): dist.all_reduce(val_loss_sum, op=dist.ReduceOp.SUM) dist.all_reduce(val_token_count, op=dist.ReduceOp.SUM) dist.all_reduce(val_byte_count, op=dist.ReduceOp.SUM) - val_loss = val_loss_sum / val_token_count bits_per_token = val_loss.item() / math.log(2.0) tokens_per_byte = val_token_count.item() / val_byte_count.item() model.train() return float(val_loss.item()), float(bits_per_token * tokens_per_byte) -# ----------------------------- -# POST-TRAINING QUANTIZATION -# ----------------------------- -# -# It's silly to export our model, which is trained in bf16 and fp32, at that same precision. -# Instead, we get approximately the same model (with a small hit) by quantizing the model to int8 & zlib compressing. -# We can then decompress the model and run in higher precision for evaluation, after closing in under the size limit. +# --- Quantization helpers --- CONTROL_TENSOR_NAME_PATTERNS = tuple( pattern for pattern in os.environ.get( "CONTROL_TENSOR_NAME_PATTERNS", - "attn_scale,attn_scales,mlp_scale,mlp_scales,resid_mix,resid_mixes,q_gain,skip_weight,skip_weights", + "attn_scale,attn_scales,mlp_scale,mlp_scales,resid_mix,resid_mixes,q_gain,skip_weight,skip_weights,skip_gates,smear,dtg_gate,ve_layer_scales,ve_shared.scale,attn_gate,vr_lambda", ).split(",") if pattern ) @@ -306,10 +401,8 @@ def eval_val( INT8_PER_ROW_SCALE_DTYPE = torch.float16 INT8_CLIP_PERCENTILE = 99.99984 INT8_CLIP_Q = INT8_CLIP_PERCENTILE / 100.0 - def tensor_nbytes(t: Tensor) -> int: return int(t.numel()) * int(t.element_size()) - def keep_float_tensor(name: str, t: Tensor, passthrough_orig_dtypes: dict[str, str]) -> Tensor: if any(pattern in name for pattern in INT8_KEEP_FLOAT_FP32_NAME_PATTERNS): return t.float().contiguous() @@ -317,12 +410,9 @@ def keep_float_tensor(name: str, t: Tensor, passthrough_orig_dtypes: dict[str, s passthrough_orig_dtypes[name] = str(t.dtype).removeprefix("torch.") return t.to(dtype=INT8_KEEP_FLOAT_STORE_DTYPE).contiguous() return t - def quantize_float_tensor(t: Tensor) -> tuple[Tensor, Tensor]: t32 = t.float() if t32.ndim == 2: - # Matrices get one scale per row, which usually tracks output-channel - # ranges much better than a single tensor-wide scale. clip_abs = ( torch.quantile(t32.abs(), INT8_CLIP_Q, dim=1) if t32.numel() @@ -332,19 +422,11 @@ def quantize_float_tensor(t: Tensor) -> tuple[Tensor, Tensor]: scale = (clip_abs / 127.0).clamp_min(1.0 / 127.0) q = torch.clamp(torch.round(clipped / scale[:, None]), -127, 127).to(torch.int8).contiguous() return q, scale.to(dtype=INT8_PER_ROW_SCALE_DTYPE).contiguous() - - # Vectors / scalars use a simpler per-tensor scale. clip_abs = float(torch.quantile(t32.abs().flatten(), INT8_CLIP_Q).item()) if t32.numel() else 0.0 scale = torch.tensor(clip_abs / 127.0 if clip_abs > 0 else 1.0, dtype=torch.float32) q = torch.clamp(torch.round(torch.clamp(t32, -clip_abs, clip_abs) / scale), -127, 127).to(torch.int8).contiguous() return q, scale - def quantize_state_dict_int8(state_dict: dict[str, Tensor]): - # Single supported clean-script export format: - # - per-row int8 for 2D float tensors - # - per-tensor int8 for other float tensors - # - exact passthrough for non-floats - # - passthrough for small float tensors, stored as fp16 to save bytes quantized: dict[str, Tensor] = {} scales: dict[str, Tensor] = {} dtypes: dict[str, str] = {} @@ -355,27 +437,21 @@ def quantize_state_dict_int8(state_dict: dict[str, Tensor]): ("param_count", "num_tensors", "num_float_tensors", "num_nonfloat_tensors", "baseline_tensor_bytes", "int8_payload_bytes"), 0, ) - for name, tensor in state_dict.items(): t = tensor.detach().to("cpu").contiguous() stats["param_count"] += int(t.numel()) stats["num_tensors"] += 1 stats["baseline_tensor_bytes"] += tensor_nbytes(t) - if not t.is_floating_point(): stats["num_nonfloat_tensors"] += 1 passthrough[name] = t stats["int8_payload_bytes"] += tensor_nbytes(t) continue - - # Small float tensors are cheap enough to keep directly. We still downcast - # fp32/bf16 passthrough tensors to fp16 so metadata does not dominate size. if t.numel() <= INT8_KEEP_FLOAT_MAX_NUMEL: kept = keep_float_tensor(name, t, passthrough_orig_dtypes) passthrough[name] = kept stats["int8_payload_bytes"] += tensor_nbytes(kept) continue - stats["num_float_tensors"] += 1 q, s = quantize_float_tensor(t) if s.ndim > 0: @@ -384,7 +460,6 @@ def quantize_state_dict_int8(state_dict: dict[str, Tensor]): scales[name] = s dtypes[name] = str(t.dtype).removeprefix("torch.") stats["int8_payload_bytes"] += tensor_nbytes(q) + tensor_nbytes(s) - obj: dict[str, object] = { "__quant_format__": "int8_clean_per_row_v1", "quantized": quantized, @@ -397,7 +472,6 @@ def quantize_state_dict_int8(state_dict: dict[str, Tensor]): if passthrough_orig_dtypes: obj["passthrough_orig_dtypes"] = passthrough_orig_dtypes return obj, stats - def dequantize_state_dict_int8(obj: dict[str, object]) -> dict[str, Tensor]: out: dict[str, Tensor] = {} qmeta = obj.get("qmeta", {}) @@ -407,13 +481,11 @@ def dequantize_state_dict_int8(obj: dict[str, object]) -> dict[str, Tensor]: s = obj["scales"][name] if qmeta.get(name, {}).get("scheme") == "per_row" or s.ndim > 0: s = s.to(dtype=torch.float32) - # Broadcast the saved row scale back across trailing dimensions. out[name] = (q.float() * s.view(q.shape[0], *([1] * (q.ndim - 1)))).to(dtype=dtype).contiguous() else: scale = float(s.item()) out[name] = (q.float() * scale).to(dtype=dtype).contiguous() for name, t in obj["passthrough"].items(): - # Restore small tensors, undoing the temporary fp16 storage cast if needed. out_t = t.detach().to("cpu").contiguous() orig_dtype = passthrough_orig_dtypes.get(name) if isinstance(orig_dtype, str): @@ -421,16 +493,12 @@ def dequantize_state_dict_int8(obj: dict[str, object]) -> dict[str, Tensor]: out[name] = out_t return out - -# ----------------------------- -# DATA LOADING -# ----------------------------- +# --- Data loading --- def load_data_shard(file: Path) -> Tensor: header_bytes = 256 * np.dtype(" Tensor: if tokens_np.size != num_tokens: raise ValueError(f"Short read for {file}") return torch.from_numpy(tokens_np.astype(np.uint16, copy=False)) - - class TokenStream: - # Reads shards sequentially and wraps around forever. The training loop therefore - # has deterministic, simple streaming behavior with no sampling or workers. def __init__(self, pattern: str): self.files = [Path(p) for p in sorted(glob.glob(pattern))] if not self.files: @@ -453,12 +517,10 @@ def __init__(self, pattern: str): self.file_idx = 0 self.tokens = load_data_shard(self.files[0]) self.pos = 0 - def _advance_file(self) -> None: self.file_idx = (self.file_idx + 1) % len(self.files) self.tokens = load_data_shard(self.files[self.file_idx]) self.pos = 0 - def take(self, n: int) -> Tensor: chunks: list[Tensor] = [] remaining = n @@ -472,17 +534,12 @@ def take(self, n: int) -> Tensor: self.pos += k remaining -= k return chunks[0] if len(chunks) == 1 else torch.cat(chunks) - - class DistributedTokenLoader: - # Each call consumes a contiguous chunk from the shared token stream, then slices out - # one disjoint span per rank. The extra "+1" token lets us build (x, y) by shifting. def __init__(self, pattern: str, rank: int, world_size: int, device: torch.device): self.rank = rank self.world_size = world_size self.device = device self.stream = TokenStream(pattern) - def next_batch(self, global_tokens: int, seq_len: int, grad_accum_steps: int) -> tuple[Tensor, Tensor]: local_tokens = global_tokens // (self.world_size * grad_accum_steps) per_rank_span = local_tokens + 1 @@ -493,44 +550,44 @@ def next_batch(self, global_tokens: int, seq_len: int, grad_accum_steps: int) -> y = local[1:].reshape(-1, seq_len) return x.to(self.device, non_blocking=True), y.to(self.device, non_blocking=True) -# ----------------------------- -# TRANSFORMER MODULES -# ----------------------------- +# --- Transformer modules --- class RMSNorm(nn.Module): def __init__(self, eps: float | None = None): super().__init__() self.eps = eps - def forward(self, x: Tensor) -> Tensor: return F.rms_norm(x, (x.size(-1),), eps=self.eps) - - class CastedLinear(nn.Linear): - # Keep weights in fp32 for optimizer/state quality, cast at matmul time for bf16 compute. + _qat_enabled: bool = False def forward(self, x: Tensor) -> Tensor: + w = self.weight.to(x.dtype) + if CastedLinear._qat_enabled and self.training and w.ndim == 2: + with torch.no_grad(): + w32 = self.weight.float() + row_max = w32.abs().amax(dim=1) + scale = (row_max / 31.0).clamp_min(1.0 / 31.0) + w_q = (torch.clamp(torch.round(w32 / scale[:, None]), -32, 31) * scale[:, None]).to(x.dtype) + w = w + (w_q - w).detach() bias = self.bias.to(x.dtype) if self.bias is not None else None - return F.linear(x, self.weight.to(x.dtype), bias) - - + return F.linear(x, w, bias) def restore_low_dim_params_to_fp32(module: nn.Module) -> None: - # Keep small/control parameters in fp32 even when the model body runs in bf16. with torch.no_grad(): for name, param in module.named_parameters(): if (param.ndim < 2 or any(pattern in name for pattern in CONTROL_TENSOR_NAME_PATTERNS)) and param.dtype != torch.float32: param.data = param.data.float() - - class Rotary(nn.Module): - # Caches cos/sin tables per sequence length on the current device. - def __init__(self, dim: int, base: float = 10000.0): + def __init__(self, dim: int, base: float = 10000.0, train_seq_len: int = 1024, rope_dims: int = 0): super().__init__() - inv_freq = 1.0 / (base ** (torch.arange(0, dim, 2, dtype=torch.float32) / dim)) + self.dim = dim + self.base = base + self.train_seq_len = train_seq_len + self.rope_dims = rope_dims if rope_dims > 0 else dim + inv_freq = 1.0 / (base ** (torch.arange(0, self.rope_dims, 2, dtype=torch.float32) / self.rope_dims)) self.register_buffer("inv_freq", inv_freq, persistent=False) self._seq_len_cached = 0 self._cos_cached: Tensor | None = None self._sin_cached: Tensor | None = None - def forward(self, seq_len: int, device: torch.device, dtype: torch.dtype) -> tuple[Tensor, Tensor]: if ( self._cos_cached is None @@ -538,20 +595,30 @@ def forward(self, seq_len: int, device: torch.device, dtype: torch.dtype) -> tup or self._seq_len_cached != seq_len or self._cos_cached.device != device ): - t = torch.arange(seq_len, device=device, dtype=self.inv_freq.dtype) - freqs = torch.outer(t, self.inv_freq.to(device)) - self._cos_cached = freqs.cos()[None, None, :, :] - self._sin_cached = freqs.sin()[None, None, :, :] + rd = self.rope_dims + if seq_len > self.train_seq_len: + scale = seq_len / self.train_seq_len + new_base = self.base * (scale ** (rd / (rd - 2))) + inv_freq = 1.0 / (new_base ** (torch.arange(0, rd, 2, dtype=torch.float32, device=device) / rd)) + else: + inv_freq = self.inv_freq.to(device) + t = torch.arange(seq_len, device=device, dtype=inv_freq.dtype) + freqs = torch.outer(t, inv_freq) + self._cos_cached = freqs.cos()[None, :, None, :] + self._sin_cached = freqs.sin()[None, :, None, :] self._seq_len_cached = seq_len return self._cos_cached.to(dtype=dtype), self._sin_cached.to(dtype=dtype) - - -def apply_rotary_emb(x: Tensor, cos: Tensor, sin: Tensor) -> Tensor: +def apply_rotary_emb(x: Tensor, cos: Tensor, sin: Tensor, rope_dims: int = 0) -> Tensor: + if rope_dims > 0 and rope_dims < x.size(-1): + x_rope, x_pass = x[..., :rope_dims], x[..., rope_dims:] + half = rope_dims // 2 + x1, x2 = x_rope[..., :half], x_rope[..., half:] + x_rope = torch.cat((x1 * cos + x2 * sin, x1 * (-sin) + x2 * cos), dim=-1) + return torch.cat((x_rope, x_pass), dim=-1) half = x.size(-1) // 2 x1, x2 = x[..., :half], x[..., half:] return torch.cat((x1 * cos + x2 * sin, x1 * (-sin) + x2 * cos), dim=-1) - class CausalSelfAttention(nn.Module): def __init__( self, @@ -560,6 +627,8 @@ def __init__( num_kv_heads: int, rope_base: float, qk_gain_init: float, + gated_attention: bool = False, + value_residual: bool = False, ): super().__init__() if dim % num_heads != 0: @@ -571,51 +640,133 @@ def __init__( self.head_dim = dim // num_heads if self.head_dim % 2 != 0: raise ValueError("head_dim must be even for RoPE") - kv_dim = self.num_kv_heads * self.head_dim - self.c_q = CastedLinear(dim, dim, bias=False) - self.c_k = CastedLinear(dim, kv_dim, bias=False) - self.c_v = CastedLinear(dim, kv_dim, bias=False) - self.proj = CastedLinear(dim, dim, bias=False) - self.proj._zero_init = True + # No CastedLinear -- weights come from banks self.q_gain = nn.Parameter(torch.full((num_heads,), qk_gain_init, dtype=torch.float32)) - self.rotary = Rotary(self.head_dim, base=rope_base) - - def forward(self, x: Tensor) -> Tensor: + self.rope_dims = 0 # set by GPT.__init__ for partial RoPE + self.rotary = Rotary(self.head_dim, base=rope_base, train_seq_len=1024) + self.use_xsa = False # set by GPT.__init__ for deep layers only + # Gated attention and value residual (non-banked small params) + self.gated_attention = gated_attention + if gated_attention: + self.attn_gate = nn.Linear(dim, num_heads, bias=True) + nn.init.zeros_(self.attn_gate.weight) + nn.init.constant_(self.attn_gate.bias, 4.0) + self.value_residual = value_residual + if value_residual: + self.vrl_alpha = nn.Parameter(torch.zeros(1, dtype=torch.float32)) # sigmoid gate (PR #569 style) + def _xsa_efficient(self, y: Tensor, v: Tensor) -> Tensor: + """Efficient XSA: subtract self-value projection via GQA-aware reshape (no repeat_interleave). + y: [B, T, H, D], v: [B, T, Hkv, D]. H must be divisible by Hkv.""" + B, T, H, D = y.shape + Hkv = v.size(-2) + group = H // Hkv + y_g = y.reshape(B, T, Hkv, group, D) # [B, T, Hkv, group, D] + vn = F.normalize(v, dim=-1).unsqueeze(-2) # [B, T, Hkv, 1, D] -- broadcast ready + proj = (y_g * vn).sum(dim=-1, keepdim=True) * vn + return (y_g - proj).reshape(B, T, H, D) + def forward(self, x: Tensor, q_w: Tensor, k_w: Tensor, v_w: Tensor, out_w: Tensor, v_embed: Tensor | None = None, v0: Tensor | None = None) -> tuple[Tensor, Tensor | None]: bsz, seqlen, dim = x.shape - q = self.c_q(x).reshape(bsz, seqlen, self.num_heads, self.head_dim).transpose(1, 2) - k = self.c_k(x).reshape(bsz, seqlen, self.num_kv_heads, self.head_dim).transpose(1, 2) - v = self.c_v(x).reshape(bsz, seqlen, self.num_kv_heads, self.head_dim).transpose(1, 2) + q = F.linear(x, q_w.to(x.dtype)).reshape(bsz, seqlen, self.num_heads, self.head_dim) + k = F.linear(x, k_w.to(x.dtype)).reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + v = F.linear(x, v_w.to(x.dtype)) + if v_embed is not None: + v = v + v_embed + v = v.reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + raw_v = v if self.value_residual else None + if self.value_residual and v0 is not None: + alpha = torch.sigmoid(self.vrl_alpha.to(dtype=v.dtype)) + v = v + alpha * v0 # sigmoid-gated residual (PR #569 style) q = F.rms_norm(q, (q.size(-1),)) k = F.rms_norm(k, (k.size(-1),)) cos, sin = self.rotary(seqlen, x.device, q.dtype) - q = apply_rotary_emb(q, cos, sin) - k = apply_rotary_emb(k, cos, sin) - q = q * self.q_gain.to(dtype=q.dtype)[None, :, None, None] - y = F.scaled_dot_product_attention( - q, - k, - v, - attn_mask=None, - is_causal=True, - enable_gqa=(self.num_kv_heads != self.num_heads), - ) - y = y.transpose(1, 2).contiguous().reshape(bsz, seqlen, dim) - return self.proj(y) + q = apply_rotary_emb(q, cos, sin, self.rope_dims) + k = apply_rotary_emb(k, cos, sin, self.rope_dims) + q = q * self.q_gain.to(dtype=q.dtype)[None, None, :, None] + if HAS_FA3: + y = flash_attn_3_func(q, k, v, causal=True) + else: + q2, k2, v2 = q.transpose(1, 2), k.transpose(1, 2), v.transpose(1, 2) + if k2.size(1) != q2.size(1): + reps = q2.size(1) // k2.size(1) + k2 = k2.repeat_interleave(reps, dim=1) + v2 = v2.repeat_interleave(reps, dim=1) + y = F.scaled_dot_product_attention(q2, k2, v2, is_causal=True).transpose(1, 2) + if self.use_xsa: + y = self._xsa_efficient(y, v) + if self.gated_attention: + # gate shape: (bsz, seqlen, num_heads) -> (bsz, seqlen, num_heads, 1) for B,T,H,D layout + gate = torch.sigmoid(self.attn_gate(x)).unsqueeze(-1) + y = y * gate + y = y.reshape(bsz, seqlen, dim) + return F.linear(y, out_w.to(x.dtype)), raw_v + +class SmearGate(nn.Module): + def __init__(self, dim: int): + super().__init__() + self.gate = nn.Parameter(torch.zeros(dim, dtype=torch.float32)) + def forward(self, x: Tensor) -> Tensor: + g = torch.sigmoid(self.gate.to(dtype=x.dtype))[None, None, :] + x_prev = torch.cat([torch.zeros_like(x[:, :1]), x[:, :-1]], dim=1) + return (1 - g) * x + g * x_prev +class BigramHashEmbedding(nn.Module): + def __init__(self, bigram_vocab_size: int, bigram_dim: int, model_dim: int, trigram: bool = False): + super().__init__() + self.bigram_vocab_size = bigram_vocab_size + self._trigram = trigram + self.embed = nn.Embedding(bigram_vocab_size, bigram_dim) + nn.init.zeros_(self.embed.weight) + self.proj = CastedLinear(bigram_dim, model_dim, bias=False) if bigram_dim != model_dim else None + if self.proj is not None: + nn.init.zeros_(self.proj.weight) + self.scale = nn.Parameter(torch.tensor(0.05, dtype=torch.float32)) + def bigram_hash(self, tokens: Tensor) -> Tensor: + t = tokens.to(torch.int32) + mod = self.bigram_vocab_size - 1 + out = torch.empty_like(t) + out[..., 0] = mod + out[..., 1:] = torch.bitwise_xor(36313 * t[..., 1:], 27191 * t[..., :-1]) % mod + return out.long() + def trigram_hash(self, tokens: Tensor) -> Tensor: + """Hash (t-2, t-1, t) trigrams into same embedding table. Zero extra params.""" + t = tokens.to(torch.int32) + mod = self.bigram_vocab_size - 1 + out = torch.empty_like(t) + out[..., :2] = mod + out[..., 2:] = (36313 * t[..., 2:] ^ 27191 * t[..., 1:-1] ^ 51497 * t[..., :-2]) % mod + return out.long() + def forward(self, token_ids: Tensor) -> Tensor: + h = self.embed(self.bigram_hash(token_ids)) + if self._trigram: + h = h + self.embed(self.trigram_hash(token_ids)) + if self.proj is not None: + h = self.proj(h) + return h * self.scale.to(dtype=h.dtype) + +class ValueEmbedding(nn.Module): + """Reinject token identity into attention values at specific layers. + Each table maps vocab tokens to a low-dim embedding, projected to model_dim.""" + def __init__(self, vocab_size: int, ve_dim: int, model_dim: int): + super().__init__() + self.embed = nn.Embedding(vocab_size, ve_dim) + nn.init.normal_(self.embed.weight, std=0.01) + self.proj = CastedLinear(ve_dim, model_dim, bias=False) if ve_dim != model_dim else None + if self.proj is not None: + nn.init.zeros_(self.proj.weight) + self.scale = nn.Parameter(torch.tensor(0.1, dtype=torch.float32)) + def forward(self, token_ids: Tensor) -> Tensor: + h = self.embed(token_ids) + if self.proj is not None: + h = self.proj(h) + return h * self.scale.to(dtype=h.dtype) class MLP(nn.Module): - # relu^2 MLP from the original modded-nanogpt setup def __init__(self, dim: int, mlp_mult: int): super().__init__() - hidden = mlp_mult * dim - self.fc = CastedLinear(dim, hidden, bias=False) - self.proj = CastedLinear(hidden, dim, bias=False) - self.proj._zero_init = True - - def forward(self, x: Tensor) -> Tensor: - x = torch.relu(self.fc(x)) - return self.proj(x.square()) - + # No CastedLinear -- weights come from banks + def forward(self, x: Tensor, up_w: Tensor, down_w: Tensor) -> Tensor: + x = F.leaky_relu(F.linear(x, up_w.to(x.dtype)), negative_slope=0.5) + return F.linear(x.square(), down_w.to(x.dtype)) class Block(nn.Module): def __init__( @@ -626,24 +777,38 @@ def __init__( mlp_mult: int, rope_base: float, qk_gain_init: float, + layer_idx: int = 0, + ln_scale: bool = False, + dtg: bool = False, + gated_attention: bool = False, + value_residual: bool = False, ): super().__init__() self.attn_norm = RMSNorm() self.mlp_norm = RMSNorm() - self.attn = CausalSelfAttention(dim, num_heads, num_kv_heads, rope_base, qk_gain_init) + self.attn = CausalSelfAttention(dim, num_heads, num_kv_heads, rope_base, qk_gain_init, + gated_attention=gated_attention, value_residual=value_residual) self.mlp = MLP(dim, mlp_mult) self.attn_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) self.mlp_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) self.resid_mix = nn.Parameter(torch.stack((torch.ones(dim), torch.zeros(dim))).float()) - - def forward(self, x: Tensor, x0: Tensor) -> Tensor: + self.ln_scale_factor = 1.0 / math.sqrt(layer_idx + 1) if ln_scale else 1.0 + if dtg: + self.dtg_gate = nn.Linear(dim, 1, bias=True) + nn.init.zeros_(self.dtg_gate.weight) + nn.init.constant_(self.dtg_gate.bias, 2.0) + else: + self.dtg_gate = None + def forward(self, x: Tensor, x0: Tensor, q_w: Tensor, k_w: Tensor, v_w: Tensor, out_w: Tensor, up_w: Tensor, down_w: Tensor, v_embed: Tensor | None = None, v0: Tensor | None = None) -> tuple[Tensor, Tensor | None]: mix = self.resid_mix.to(dtype=x.dtype) - x = mix[0][None, None, :] * x + mix[1][None, None, :] * x0 - attn_out = self.attn(self.attn_norm(x)) - x = x + self.attn_scale.to(dtype=x.dtype)[None, None, :] * attn_out - x = x + self.mlp_scale.to(dtype=x.dtype)[None, None, :] * self.mlp(self.mlp_norm(x)) - return x - + x_in = mix[0][None, None, :] * x + mix[1][None, None, :] * x0 + attn_out, raw_v = self.attn(self.attn_norm(x_in) * self.ln_scale_factor, q_w, k_w, v_w, out_w, v_embed=v_embed, v0=v0) + x_out = x_in + self.attn_scale.to(dtype=x_in.dtype)[None, None, :] * attn_out + x_out = x_out + self.mlp_scale.to(dtype=x_out.dtype)[None, None, :] * self.mlp(self.mlp_norm(x_out) * self.ln_scale_factor, up_w, down_w) + if self.dtg_gate is not None: + gate = torch.sigmoid(self.dtg_gate(x_in.detach())) + x_out = x_in + gate * (x_out - x_in) + return x_out, raw_v class GPT(nn.Module): def __init__( @@ -659,18 +824,48 @@ def __init__( logit_softcap: float, rope_base: float, qk_gain_init: float, + mtp_num_heads: int = 0, + mtp_loss_weight: float = 0.1, + bigram_vocab_size: int = 0, + bigram_dim: int = 128, + xsa_last_n: int = 0, + rope_dims: int = 0, + ln_scale: bool = False, + dtg: bool = False, + ve_enabled: bool = False, + ve_dim: int = 128, + ve_layers: str = "9,10", + gated_attention: bool = False, + value_residual: bool = False, + skip_gates_enabled: bool = True, ): super().__init__() + self._ve_target_dim = num_kv_heads * (model_dim // num_heads) # kv_dim for value projection if logit_softcap <= 0.0: raise ValueError(f"logit_softcap must be positive, got {logit_softcap}") self.tie_embeddings = tie_embeddings self.tied_embed_init_std = tied_embed_init_std self.logit_softcap = logit_softcap + self.value_residual = value_residual + self.mtp_num_heads = mtp_num_heads + self.mtp_loss_weight = mtp_loss_weight self.tok_emb = nn.Embedding(vocab_size, model_dim) + self.bigram = BigramHashEmbedding(bigram_vocab_size, bigram_dim, model_dim, trigram=bool(int(os.environ.get("TRIGRAM", "0")))) if bigram_vocab_size > 0 else None + self.smear = SmearGate(model_dim) self.num_encoder_layers = num_layers // 2 self.num_decoder_layers = num_layers - self.num_encoder_layers self.num_skip_weights = min(self.num_encoder_layers, self.num_decoder_layers) self.skip_weights = nn.Parameter(torch.ones(self.num_skip_weights, model_dim, dtype=torch.float32)) + self.skip_gates = nn.Parameter(torch.zeros(self.num_skip_weights, model_dim, dtype=torch.float32)) if skip_gates_enabled else None + # Parameter banks: contiguous 3D tensors for batched optimizer + head_dim = model_dim // num_heads + kv_dim = num_kv_heads * head_dim + mlp_dim = int(mlp_mult * model_dim) + self.num_layers = num_layers + self.qo_bank = nn.Parameter(torch.empty(2 * num_layers, model_dim, model_dim)) + self.kv_bank = nn.Parameter(torch.empty(2 * num_layers, kv_dim, model_dim)) + self.mlp_up_bank = nn.Parameter(torch.empty(num_layers, mlp_dim, model_dim)) + self.mlp_down_bank = nn.Parameter(torch.empty(num_layers, model_dim, mlp_dim)) self.blocks = nn.ModuleList( [ Block( @@ -680,65 +875,853 @@ def __init__( mlp_mult, rope_base, qk_gain_init, + layer_idx=i, + ln_scale=ln_scale, + dtg=dtg, + gated_attention=gated_attention, + value_residual=value_residual, ) for i in range(num_layers) ] ) + if rope_dims > 0: + head_dim = model_dim // num_heads + for block in self.blocks: + block.attn.rope_dims = rope_dims + block.attn.rotary = Rotary(head_dim, base=rope_base, train_seq_len=1024, rope_dims=rope_dims) + self.ve_layer_indices = [int(x) for x in ve_layers.split(",") if x.strip()] if ve_enabled else [] + kv_dim_ve = self._ve_target_dim + if self.ve_layer_indices: + self.ve_shared = ValueEmbedding(vocab_size, ve_dim, kv_dim_ve) + self.ve_layer_scales = nn.ParameterList( + [nn.Parameter(torch.ones(1, dtype=torch.float32)) for _ in self.ve_layer_indices] + ) + else: + self.ve_shared = None + self.ve_layer_scales = nn.ParameterList() + self.value_embeds = nn.ModuleList() # keep empty for compat self.final_norm = RMSNorm() self.lm_head = None if tie_embeddings else CastedLinear(model_dim, vocab_size, bias=False) if self.lm_head is not None: self.lm_head._zero_init = True + self.mtp_heads = nn.ModuleList( + [CastedLinear(model_dim, vocab_size, bias=False) for _ in range(mtp_num_heads)] + ) + for head in self.mtp_heads: + head._zero_init = True + if xsa_last_n > 0: + for i in range(max(0, num_layers - xsa_last_n), num_layers): + self.blocks[i].attn.use_xsa = True self._init_weights() - def _init_weights(self) -> None: if self.tie_embeddings: nn.init.normal_(self.tok_emb.weight, mean=0.0, std=self.tied_embed_init_std) - for module in self.modules(): - if isinstance(module, nn.Linear) and getattr(module, "_zero_init", False): - nn.init.zeros_(module.weight) - + n = self.num_layers + proj_scale = 1.0 / math.sqrt(2 * n) + # Init banks: orthogonal, with proj layers scaled down and out/down zero-init + for i in range(n): + nn.init.orthogonal_(self.qo_bank.data[i], gain=1.0) # Q + nn.init.zeros_(self.qo_bank.data[n + i]) # Out (zero init) + nn.init.orthogonal_(self.kv_bank.data[i], gain=1.0) # K + nn.init.orthogonal_(self.kv_bank.data[n + i], gain=1.0) # V + nn.init.orthogonal_(self.mlp_up_bank.data[i], gain=1.0) # MLP up + nn.init.zeros_(self.mlp_down_bank.data[i]) # MLP down (zero init) + # Scale proj layers (out_proj and mlp_down are "proj" layers) + self.qo_bank.data[n + i].mul_(proj_scale) + self.mlp_down_bank.data[i].mul_(proj_scale) + # Init remaining nn.Linear modules (bigram proj, mtp heads, lm_head) + for name, module in self.named_modules(): + if isinstance(module, nn.Linear): + if getattr(module, "_zero_init", False): + nn.init.zeros_(module.weight) + elif module.weight.ndim == 2 and module.weight.shape[0] >= 64 and module.weight.shape[1] >= 64: + nn.init.orthogonal_(module.weight, gain=1.0) + def _get_ve(self, layer_idx: int, input_ids: Tensor, ve_cache: dict | None = None) -> Tensor | None: + """Get value embedding for a specific layer using shared table + per-layer scale.""" + if self.ve_shared is None or layer_idx not in self.ve_layer_indices: + return None + if ve_cache is not None and 've' not in ve_cache: + ve_cache['ve'] = self.ve_shared(input_ids) + ve_base = ve_cache['ve'] if ve_cache is not None else self.ve_shared(input_ids) + ve_idx = self.ve_layer_indices.index(layer_idx) + return ve_base * self.ve_layer_scales[ve_idx].to(dtype=ve_base.dtype) def forward(self, input_ids: Tensor, target_ids: Tensor) -> Tensor: + n = self.num_layers x = self.tok_emb(input_ids) + if self.bigram is not None: + x = x + self.bigram(input_ids) x = F.rms_norm(x, (x.size(-1),)) + x = self.smear(x) x0 = x + v0 = None skips: list[Tensor] = [] - - # First half stores skips; second half reuses them in reverse order. + ve_cache: dict = {} for i in range(self.num_encoder_layers): - x = self.blocks[i](x, x0) + ve = self._get_ve(i, input_ids, ve_cache) + x, raw_v = self.blocks[i](x, x0, + self.qo_bank[i], self.kv_bank[i], self.kv_bank[n + i], + self.qo_bank[n + i], self.mlp_up_bank[i], self.mlp_down_bank[i], + v_embed=ve, v0=v0) + if v0 is None and raw_v is not None: + v0 = raw_v skips.append(x) for i in range(self.num_decoder_layers): + bi = self.num_encoder_layers + i if skips: - x = x + self.skip_weights[i].to(dtype=x.dtype)[None, None, :] * skips.pop() - x = self.blocks[self.num_encoder_layers + i](x, x0) - - x = self.final_norm(x).reshape(-1, x.size(-1)) + scaled_skip = self.skip_weights[i].to(dtype=x.dtype)[None, None, :] * skips.pop() + if self.skip_gates is not None: + g = torch.sigmoid(self.skip_gates[i].to(dtype=x.dtype))[None, None, :] + x = torch.lerp(scaled_skip, x, g) + else: + x = x + scaled_skip + ve = self._get_ve(bi, input_ids, ve_cache) + x, _ = self.blocks[bi](x, x0, + self.qo_bank[bi], self.kv_bank[bi], self.kv_bank[n + bi], + self.qo_bank[n + bi], self.mlp_up_bank[bi], self.mlp_down_bank[bi], + v_embed=ve, v0=v0) + x = self.final_norm(x) + x_flat = x.reshape(-1, x.size(-1)) targets = target_ids.reshape(-1) if self.tie_embeddings: - logits_proj = F.linear(x, self.tok_emb.weight) + logits_proj = F.linear(x_flat, self.tok_emb.weight) else: if self.lm_head is None: raise RuntimeError("lm_head is required when tie_embeddings=False") - logits_proj = self.lm_head(x) + logits_proj = self.lm_head(x_flat) + logits = self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + main_loss = F.cross_entropy(logits.float(), targets, reduction="mean") + if self.training and self.mtp_num_heads > 0 and self.mtp_loss_weight > 0.0: + _, seqlen, dim = x.shape + mtp_loss_sum = x.new_zeros(()) + mtp_loss_count = 0 + for k, mtp_head in enumerate(self.mtp_heads): + valid_t = seqlen - (k + 1) + if valid_t <= 0: + continue + mtp_hidden = x[:, :valid_t, :].reshape(-1, dim) + mtp_targets = target_ids[:, k + 1 :].reshape(-1) + mtp_logits_proj = mtp_head(mtp_hidden) + mtp_logits = self.logit_softcap * torch.tanh(mtp_logits_proj / self.logit_softcap) + mtp_loss_sum = mtp_loss_sum + F.cross_entropy(mtp_logits.float(), mtp_targets, reduction="mean") + mtp_loss_count += 1 + if mtp_loss_count > 0: + main_loss = main_loss + self.mtp_loss_weight * (mtp_loss_sum / mtp_loss_count) + return main_loss + def forward_hidden(self, input_ids: Tensor) -> Tensor: + """Return final hidden states before lm_head.""" + n = self.num_layers + x = self.tok_emb(input_ids) + if self.bigram is not None: + x = x + self.bigram(input_ids) + x = F.rms_norm(x, (x.size(-1),)) + x = self.smear(x) + x0 = x + v0 = None + skips: list[Tensor] = [] + ve_cache: dict = {} + for i in range(self.num_encoder_layers): + ve = self._get_ve(i, input_ids, ve_cache) + x, raw_v = self.blocks[i](x, x0, + self.qo_bank[i], self.kv_bank[i], self.kv_bank[n + i], + self.qo_bank[n + i], self.mlp_up_bank[i], self.mlp_down_bank[i], + v_embed=ve, v0=v0) + if v0 is None and raw_v is not None: + v0 = raw_v + skips.append(x) + for i in range(self.num_decoder_layers): + bi = self.num_encoder_layers + i + if skips: + scaled_skip = self.skip_weights[i].to(dtype=x.dtype)[None, None, :] * skips.pop() + if self.skip_gates is not None: + g = torch.sigmoid(self.skip_gates[i].to(dtype=x.dtype))[None, None, :] + x = torch.lerp(scaled_skip, x, g) + else: + x = x + scaled_skip + ve = self._get_ve(bi, input_ids, ve_cache) + x, _ = self.blocks[bi](x, x0, + self.qo_bank[bi], self.kv_bank[bi], self.kv_bank[n + bi], + self.qo_bank[n + bi], self.mlp_up_bank[bi], self.mlp_down_bank[bi], + v_embed=ve, v0=v0) + return self.final_norm(x) + + def compute_logits(self, hidden_states: Tensor) -> Tensor: + """Compute logits from hidden states.""" + if self.tie_embeddings: + logits_proj = F.linear(hidden_states, self.tok_emb.weight) + else: + logits_proj = self.lm_head(hidden_states) + return self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + + def forward_logits(self, input_ids: Tensor) -> Tensor: + """Return logits (bsz, seq_len, vocab) without computing loss.""" + return self.compute_logits(self.forward_hidden(input_ids)) + +# --- Sliding window evaluation --- + +def eval_val_sliding( + args: Hyperparameters, + base_model: nn.Module, + rank: int, + world_size: int, + device: torch.device, + val_tokens: Tensor, + base_bytes_lut: Tensor, + has_leading_space_lut: Tensor, + is_boundary_token_lut: Tensor, + stride: int, + batch_seqs: int = 32, + eval_seq_len: int | None = None, +) -> tuple[float, float]: + """Sliding window evaluation: each token scored with maximum context.""" + seq_len = eval_seq_len or args.train_seq_len + total_tokens = val_tokens.numel() - 1 + window_starts = [ws for ws in range(0, total_tokens, stride) + if min(ws + seq_len, total_tokens) - ws >= 1] + total_windows = len(window_starts) + my_s = (total_windows * rank) // world_size + my_e = (total_windows * (rank + 1)) // world_size + my_windows = window_starts[my_s:my_e] + loss_sum = torch.zeros((), device=device, dtype=torch.float64) + token_count = torch.zeros((), device=device, dtype=torch.float64) + byte_count = torch.zeros((), device=device, dtype=torch.float64) + base_model.eval() + compiled_logits = torch.compile(base_model.forward_logits, dynamic=False, fullgraph=True) + with torch.inference_mode(): + for bi in range(0, len(my_windows), batch_seqs): + batch_ws = my_windows[bi:bi + batch_seqs] + bsz = len(batch_ws) + x_batch = torch.zeros(bsz, seq_len, dtype=torch.int64, device=device) + y_batch = torch.zeros(bsz, seq_len, dtype=torch.int64, device=device) + wlens: list[int] = [] + for i, ws in enumerate(batch_ws): + end = min(ws + seq_len, total_tokens) + wlen = end - ws + wlens.append(wlen) + chunk = val_tokens[ws:end + 1].to(dtype=torch.int64, device=device) + x_batch[i, :wlen] = chunk[:-1] + y_batch[i, :wlen] = chunk[1:] + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + logits = compiled_logits(x_batch) + nll = F.cross_entropy( + logits.reshape(-1, logits.size(-1)).float(), + y_batch.reshape(-1), + reduction="none", + ).reshape(bsz, seq_len) + for i, ws in enumerate(batch_ws): + wlen = wlens[i] + s = 0 if ws == 0 else max(wlen - stride, 0) + scored_nll = nll[i, s:wlen].to(torch.float64) + loss_sum += scored_nll.sum() + token_count += float(wlen - s) + tgt = y_batch[i, s:wlen] + prev = x_batch[i, s:wlen] + tb = base_bytes_lut[tgt].to(torch.float64) + tb += (has_leading_space_lut[tgt] & ~is_boundary_token_lut[prev]).to(torch.float64) + byte_count += tb.sum() + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(token_count, op=dist.ReduceOp.SUM) + dist.all_reduce(byte_count, op=dist.ReduceOp.SUM) + val_loss = (loss_sum / token_count).item() + bits_per_token = val_loss / math.log(2.0) + tokens_per_byte = token_count.item() / byte_count.item() + base_model.train() + return val_loss, bits_per_token * tokens_per_byte + + +def ttt_adapt_adamw( + args, base_model: nn.Module, device: torch.device, + val_tokens: Tensor, rank: int = 0, world_size: int = 1, log0=print, +) -> None: + """AdamW TTT: fine-tune on val data BEFORE quantization (PR #1006 style).""" + seq_len = args.train_seq_len + total_seqs = (val_tokens.numel() - 1) // seq_len + batch_seqs = args.ttt_batch_seqs + if args.ttt_freeze_blocks > 0: + for i, block in enumerate(base_model.blocks): + if i < args.ttt_freeze_blocks: + for p in block.parameters(): + p.requires_grad_(False) + ttt_params = [p for p in base_model.parameters() if p.requires_grad] + log0(f"ttt_adamw:params trainable={sum(p.numel() for p in ttt_params)} " + f"frozen={sum(p.numel() for p in base_model.parameters() if not p.requires_grad)}") + optimizer = torch.optim.AdamW(ttt_params, lr=args.ttt_lr, weight_decay=0.0) + scheduler = None + if args.ttt_cosine_decay: + scheduler = torch.optim.lr_scheduler.CosineAnnealingLR( + optimizer, T_max=args.ttt_epochs, eta_min=args.ttt_lr * 0.1) + my_start = (total_seqs * rank) // world_size + my_end = (total_seqs * (rank + 1)) // world_size + base_model.train() + t0 = time.perf_counter() + for epoch in range(args.ttt_epochs): + epoch_loss_sum = torch.zeros((), device=device, dtype=torch.float64) + epoch_tokens = torch.zeros((), device=device, dtype=torch.float64) + for bs in range(my_start, my_end, batch_seqs): + be = min(bs + batch_seqs, my_end) + raw_start = bs * seq_len + raw_end = be * seq_len + 1 + if raw_end > val_tokens.numel(): + continue + local = val_tokens[raw_start:raw_end].to(device=device, dtype=torch.int64) + x = local[:-1].reshape(-1, seq_len) + y = local[1:].reshape(-1, seq_len) + optimizer.zero_grad(set_to_none=True) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + loss = base_model(x, y) + loss.backward() + if world_size > 1: + for p in ttt_params: + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG) + torch.nn.utils.clip_grad_norm_(ttt_params, args.ttt_grad_clip) + optimizer.step() + epoch_loss_sum += loss.detach().to(torch.float64) * float(y.numel()) + epoch_tokens += float(y.numel()) + if world_size > 1: + dist.all_reduce(epoch_loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(epoch_tokens, op=dist.ReduceOp.SUM) + epoch_avg_loss = epoch_loss_sum.item() / max(epoch_tokens.item(), 1) + if scheduler is not None: + scheduler.step() + log0(f"ttt_adamw:epoch {epoch+1}/{args.ttt_epochs} loss:{epoch_avg_loss:.4f} " + f"time:{time.perf_counter() - t0:.1f}s") + for p in base_model.parameters(): + p.requires_grad_(True) + base_model.eval() + log0(f"ttt_adamw:done elapsed={time.perf_counter() - t0:.1f}s") + + +def generate_autoregressive_calib(model, device, num_seqs=64, seq_len=2048, + vocab_size=1024, temperature=0.8, batch_size=8, seed=42): + """Generate sequences autoregressively from the model for GPTQ calibration. + No external data accessed — fully self-contained.""" + model.eval() + rng = torch.Generator(device=device) + rng.manual_seed(seed) + all_tokens = [] + with torch.inference_mode(), torch.autocast(device_type="cuda", dtype=torch.bfloat16): + for batch_start in range(0, num_seqs, batch_size): + bs = min(batch_size, num_seqs - batch_start) + tokens = torch.randint(0, vocab_size, (bs, 1), device=device, generator=rng) + for pos in range(seq_len - 1): + logits = model.forward_logits(tokens) + next_logit = logits[:, -1, :] + probs = torch.softmax(next_logit / temperature, dim=-1) + next_tok = torch.multinomial(probs, 1, generator=rng) + tokens = torch.cat([tokens, next_tok], dim=1) + for i in range(bs): + all_tokens.append(tokens[i:i+1]) + return all_tokens + + +def collect_hessians_from_tokens(hessian_model, token_seqs, device): + """Collect H = X^T X from pre-generated token sequences.""" + hessians = {} + hooks = [] + for name, module in hessian_model.named_modules(): + if isinstance(module, CastedLinear): + param_name = name + ".weight" + cols = module.weight.shape[1] + hessians[param_name] = torch.zeros(cols, cols, dtype=torch.float32, device='cpu') + h = module.register_forward_hook(_make_input_hook(hessians, param_name)) + hooks.append(h) + hessian_model.eval() + with torch.inference_mode(), torch.autocast(device_type="cuda", dtype=torch.bfloat16): + for seq in token_seqs: + x = seq[:, :-1].to(device) + y = seq[:, 1:].to(device) + hessian_model(x, y) + for h in hooks: + h.remove() + num_batches = len(token_seqs) + for name in hessians: + H = hessians[name] + H /= num_batches + damp = 0.01 * torch.diag(H).mean().clamp_min(1e-6) + H += damp * torch.eye(H.shape[0]) + hessians[name] = H + return hessians + + +# --- GPTQ-lite int6 quantization --- + +def _classify_param(name: str) -> str: + if "tok_emb" in name or "lm_head" in name: + return "embed" + if ".mlp." in name: + return "mlp" + if ".attn." in name or (".proj." in name and ".mlp." not in name): + return "attn" + return "other" +def sdclip_scale(t32: Tensor, clip_sigmas: float, clip_range: int) -> Tensor: + """SDClip: scale = clip_sigmas * std(row) / clip_range (per-row for 2D, global for 1D).""" + if t32.ndim == 2: + row_std = t32.std(dim=1).clamp_min(1e-10) + return (clip_sigmas * row_std / clip_range).clamp_min(1.0 / clip_range).to(torch.float16) + amax = t32.abs().max().item() + return torch.tensor(amax / clip_range if amax > 0 else 1.0, dtype=torch.float16) + +def quantize_int6_per_row(t: Tensor, clip_range: int = 31, clip_sigmas: float = 12.85) -> tuple[Tensor, Tensor]: + t32 = t.float() + s = sdclip_scale(t32, clip_sigmas, clip_range) + if t32.ndim == 2: + q = torch.clamp(torch.round(t32 / s.float()[:, None]), -clip_range, clip_range).to(torch.int8) + else: + q = torch.clamp(torch.round(t32 / s.float()), -clip_range, clip_range).to(torch.int8) + return q, s + +def quantize_int6_gptq(weight, hessian=None, clip_range=31, block_size=128, clip_sigmas=12.85): + """Full GPTQ: Hessian-aware int6 quantization with Cholesky error compensation. + Uses SDClip (k * std per row) for scale computation. + If hessian is None, falls back to SDClip-only quantization.""" + t32 = weight.float() + if t32.ndim != 2 or hessian is None: + return quantize_int6_per_row(t32, clip_range=clip_range, clip_sigmas=clip_sigmas) + rows, cols = t32.shape + H = hessian.float().clone() + dead = torch.diag(H) == 0 + H[dead, dead] = 1 + damp = 0.01 * torch.mean(torch.diag(H)) + H[torch.arange(cols), torch.arange(cols)] += damp + perm = torch.argsort(torch.diag(H), descending=True) + inv_perm = torch.argsort(perm) + W = t32[:, perm].clone() + W[:, dead[perm]] = 0 + H = H[perm][:, perm] + Hinv = torch.linalg.cholesky(H) + Hinv = torch.cholesky_inverse(Hinv) + Hinv = torch.linalg.cholesky(Hinv, upper=True) + # SDClip: single-pass scale from k * std(row) + s = sdclip_scale(t32, clip_sigmas, clip_range) + sf = s.float() + Q = torch.zeros_like(W, dtype=torch.int8) + W_work = W.clone() + for i1 in range(0, cols, block_size): + i2 = min(i1 + block_size, cols) + count = i2 - i1 + W1 = W_work[:, i1:i2].clone() + Q1 = torch.zeros(rows, count, dtype=torch.int8) + Err1 = torch.zeros(rows, count) + Hinv1 = Hinv[i1:i2, i1:i2] + for i in range(count): + w = W1[:, i] + d = Hinv1[i, i] + q = torch.clamp(torch.round(w / sf), -clip_range, clip_range).to(torch.int8) + Q1[:, i] = q + err = (w - q.float() * sf) / d + W1[:, i:] -= err.unsqueeze(1) * Hinv1[i, i:].unsqueeze(0) + Err1[:, i] = err + Q[:, i1:i2] = Q1 + if i2 < cols: + W_work[:, i2:] -= Err1 @ Hinv[i1:i2, i2:] + Q = Q[:, inv_perm] + return Q, s + +def _quantize_int6_percentile(t32, clip_range=31, clip_sigmas=12.85): + """Fallback: SDClip-based (kept for API compat, now uses SDClip not percentile).""" + return quantize_int6_per_row(t32, clip_range=clip_range, clip_sigmas=clip_sigmas) + +def _unbank_state_dict(sd: dict[str, Tensor], num_layers: int) -> dict[str, Tensor]: + """Convert 3D bank tensors into individual 2D tensors with standard names.""" + out: dict[str, Tensor] = {} + n = num_layers + for name, tensor in sd.items(): + if name == "qo_bank": + for i in range(n): + out[f"blocks.{i}.attn.c_q.weight"] = tensor[i] + out[f"blocks.{i}.attn.proj.weight"] = tensor[n + i] + elif name == "kv_bank": + for i in range(n): + out[f"blocks.{i}.attn.c_k.weight"] = tensor[i] + out[f"blocks.{i}.attn.c_v.weight"] = tensor[n + i] + elif name == "mlp_up_bank": + for i in range(n): + out[f"blocks.{i}.mlp.fc.weight"] = tensor[i] + elif name == "mlp_down_bank": + for i in range(n): + out[f"blocks.{i}.mlp.proj.weight"] = tensor[i] + else: + out[name] = tensor + return out + +def _rebank_state_dict(sd: dict[str, Tensor], num_layers: int, template_sd: dict[str, Tensor]) -> dict[str, Tensor]: + """Convert individual 2D tensors back into 3D bank tensors.""" + out: dict[str, Tensor] = {} + n = num_layers + # Reconstruct banks from individual weight keys + qo_slices = [None] * (2 * n) + kv_slices = [None] * (2 * n) + up_slices = [None] * n + down_slices = [None] * n + consumed = set() + for i in range(n): + qk = f"blocks.{i}.attn.c_q.weight" + if qk in sd: + qo_slices[i] = sd[qk] + consumed.add(qk) + ok = f"blocks.{i}.attn.proj.weight" + if ok in sd: + qo_slices[n + i] = sd[ok] + consumed.add(ok) + kk = f"blocks.{i}.attn.c_k.weight" + if kk in sd: + kv_slices[i] = sd[kk] + consumed.add(kk) + vk = f"blocks.{i}.attn.c_v.weight" + if vk in sd: + kv_slices[n + i] = sd[vk] + consumed.add(vk) + fk = f"blocks.{i}.mlp.fc.weight" + if fk in sd: + up_slices[i] = sd[fk] + consumed.add(fk) + dk = f"blocks.{i}.mlp.proj.weight" + if dk in sd: + down_slices[i] = sd[dk] + consumed.add(dk) + out["qo_bank"] = torch.stack(qo_slices).to(dtype=template_sd["qo_bank"].dtype) + out["kv_bank"] = torch.stack(kv_slices).to(dtype=template_sd["kv_bank"].dtype) + out["mlp_up_bank"] = torch.stack(up_slices).to(dtype=template_sd["mlp_up_bank"].dtype) + out["mlp_down_bank"] = torch.stack(down_slices).to(dtype=template_sd["mlp_down_bank"].dtype) + for name, tensor in sd.items(): + if name not in consumed: + out[name] = tensor + return out + +# --- Non-banked model for Hessian collection --- +# This mirrors the unbanked state dict keys: blocks.{i}.attn.c_q/c_k/c_v/proj, blocks.{i}.mlp.fc/proj + +class _HessianAttn(nn.Module): + """Non-banked attention with CastedLinear layers for Hessian hooks.""" + def __init__(self, dim, num_heads, num_kv_heads, rope_base, qk_gain_init): + super().__init__() + self.num_heads, self.num_kv_heads = num_heads, num_kv_heads + self.head_dim = dim // num_heads + kv_dim = num_kv_heads * self.head_dim + self.c_q = CastedLinear(dim, dim, bias=False) + self.c_k = CastedLinear(dim, kv_dim, bias=False) + self.c_v = CastedLinear(dim, kv_dim, bias=False) + self.proj = CastedLinear(dim, dim, bias=False) + self.q_gain = nn.Parameter(torch.full((num_heads,), qk_gain_init, dtype=torch.float32)) + self.rope_dims = 0 + self.rotary = Rotary(self.head_dim, base=rope_base, train_seq_len=1024) + self.use_xsa = False + def _xsa_efficient(self, y, v): + B, T, H, D = y.shape; Hkv = v.size(-2); group = H // Hkv + y_g = y.reshape(B, T, Hkv, group, D) + vn = F.normalize(v, dim=-1).unsqueeze(-2) + proj = (y_g * vn).sum(dim=-1, keepdim=True) * vn + return (y_g - proj).reshape(B, T, H, D) + def forward(self, x, v_embed=None): + bsz, seqlen, dim = x.shape + q = self.c_q(x).reshape(bsz, seqlen, self.num_heads, self.head_dim) + k = self.c_k(x).reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + v = self.c_v(x) + if v_embed is not None: + v = v + v_embed + v = v.reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + q = F.rms_norm(q, (q.size(-1),)) + k = F.rms_norm(k, (k.size(-1),)) + cos, sin = self.rotary(seqlen, x.device, q.dtype) + q = apply_rotary_emb(q, cos, sin, self.rope_dims) + k = apply_rotary_emb(k, cos, sin, self.rope_dims) + q = q * self.q_gain.to(dtype=q.dtype)[None, None, :, None] + if HAS_FA3: + y = flash_attn_3_func(q, k, v, causal=True) + else: + q2, k2, v2 = q.transpose(1, 2), k.transpose(1, 2), v.transpose(1, 2) + if k2.size(1) != q2.size(1): + reps = q2.size(1) // k2.size(1) + k2 = k2.repeat_interleave(reps, dim=1) + v2 = v2.repeat_interleave(reps, dim=1) + y = F.scaled_dot_product_attention(q2, k2, v2, is_causal=True).transpose(1, 2) + if self.use_xsa: + y = self._xsa_efficient(y, v) + return self.proj(y.reshape(bsz, seqlen, dim)) + +class _HessianMLP(nn.Module): + """Non-banked MLP with CastedLinear layers for Hessian hooks.""" + def __init__(self, dim, mlp_mult): + super().__init__() + self.fc = CastedLinear(dim, int(mlp_mult * dim), bias=False) + self.proj = CastedLinear(int(mlp_mult * dim), dim, bias=False) + def forward(self, x): + return self.proj(F.leaky_relu(self.fc(x), negative_slope=0.5).square()) + +class _HessianBlock(nn.Module): + def __init__(self, dim, num_heads, num_kv_heads, mlp_mult, rope_base, qk_gain_init, layer_idx=0, ln_scale=False): + super().__init__() + self.attn_norm = RMSNorm() + self.mlp_norm = RMSNorm() + self.attn = _HessianAttn(dim, num_heads, num_kv_heads, rope_base, qk_gain_init) + self.mlp = _HessianMLP(dim, mlp_mult) + self.attn_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.mlp_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.resid_mix = nn.Parameter(torch.stack((torch.ones(dim), torch.zeros(dim))).float()) + self.ln_scale_factor = 1.0 / math.sqrt(layer_idx + 1) if ln_scale else 1.0 + def forward(self, x, x0, v_embed=None): + mix = self.resid_mix.to(dtype=x.dtype) + x_in = mix[0][None, None, :] * x + mix[1][None, None, :] * x0 + attn_out = self.attn(self.attn_norm(x_in) * self.ln_scale_factor, v_embed=v_embed) + x_out = x_in + self.attn_scale.to(dtype=x_in.dtype)[None, None, :] * attn_out + x_out = x_out + self.mlp_scale.to(dtype=x_out.dtype)[None, None, :] * self.mlp(self.mlp_norm(x_out) * self.ln_scale_factor) + return x_out + +class _HessianGPT(nn.Module): + """Non-banked GPT model matching unbanked state dict keys for Hessian collection.""" + def __init__(self, vocab_size, num_layers, model_dim, num_heads, num_kv_heads, + mlp_mult, tie_embeddings, logit_softcap, rope_base, qk_gain_init, + bigram_vocab_size=0, bigram_dim=128, xsa_last_n=0, + rope_dims=0, ln_scale=False, + ve_enabled=False, ve_dim=128, ve_layers="9,10"): + super().__init__() + self.tie_embeddings = tie_embeddings + self.logit_softcap = logit_softcap + self.num_layers = num_layers + self.tok_emb = nn.Embedding(vocab_size, model_dim) + self.bigram = BigramHashEmbedding(bigram_vocab_size, bigram_dim, model_dim, trigram=bool(int(os.environ.get("TRIGRAM", "0")))) if bigram_vocab_size > 0 else None + self.smear = SmearGate(model_dim) + self.num_encoder_layers = num_layers // 2 + self.num_decoder_layers = num_layers - self.num_encoder_layers + self.num_skip_weights = min(self.num_encoder_layers, self.num_decoder_layers) + self.skip_weights = nn.Parameter(torch.ones(self.num_skip_weights, model_dim, dtype=torch.float32)) + self.blocks = nn.ModuleList([ + _HessianBlock(model_dim, num_heads, num_kv_heads, mlp_mult, rope_base, qk_gain_init, + layer_idx=i, ln_scale=ln_scale) + for i in range(num_layers) + ]) + if rope_dims > 0: + head_dim = model_dim // num_heads + for block in self.blocks: + block.attn.rope_dims = rope_dims + block.attn.rotary = Rotary(head_dim, base=rope_base, train_seq_len=1024, rope_dims=rope_dims) + if xsa_last_n > 0: + for i in range(max(0, num_layers - xsa_last_n), num_layers): + self.blocks[i].attn.use_xsa = True + kv_dim = num_kv_heads * (model_dim // num_heads) + self.ve_layer_indices = [int(x) for x in ve_layers.split(",") if x.strip()] if ve_enabled else [] + if self.ve_layer_indices: + self.ve_shared = ValueEmbedding(vocab_size, ve_dim, kv_dim) + self.ve_layer_scales = nn.ParameterList([nn.Parameter(torch.ones(1, dtype=torch.float32)) for _ in self.ve_layer_indices]) + else: + self.ve_shared = None + self.ve_layer_scales = nn.ParameterList() + self.final_norm = RMSNorm() + self.lm_head = None if tie_embeddings else CastedLinear(model_dim, vocab_size, bias=False) + def _get_ve(self, layer_idx, input_ids, ve_cache): + if self.ve_shared is None or layer_idx not in self.ve_layer_indices: + return None + if 've' not in ve_cache: + ve_cache['ve'] = self.ve_shared(input_ids) + ve_idx = self.ve_layer_indices.index(layer_idx) + return ve_cache['ve'] * self.ve_layer_scales[ve_idx].to(dtype=ve_cache['ve'].dtype) + def forward(self, input_ids, target_ids): + x = self.tok_emb(input_ids) + if self.bigram is not None: + x = x + self.bigram(input_ids) + x = F.rms_norm(x, (x.size(-1),)) + x = self.smear(x) + x0 = x + skips = [] + ve_cache = {} + for i in range(self.num_encoder_layers): + ve = self._get_ve(i, input_ids, ve_cache) + x = self.blocks[i](x, x0, v_embed=ve) + skips.append(x) + for i in range(self.num_decoder_layers): + bi = self.num_encoder_layers + i + if skips: + x = x + self.skip_weights[i].to(dtype=x.dtype)[None, None, :] * skips.pop() + ve = self._get_ve(bi, input_ids, ve_cache) + x = self.blocks[bi](x, x0, v_embed=ve) + x = self.final_norm(x) + x_flat = x.reshape(-1, x.size(-1)) + targets = target_ids.reshape(-1) + logits_proj = F.linear(x_flat, self.tok_emb.weight) if self.tie_embeddings else self.lm_head(x_flat) logits = self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) return F.cross_entropy(logits.float(), targets, reduction="mean") +def _make_input_hook(hessians, pname): + def hook_fn(module, input, output): + x = input[0].detach().float() + if x.ndim == 3: + x = x.reshape(-1, x.shape[-1]) + hessians[pname] += (x.T @ x).cpu() + return hook_fn + +def _make_output_hook(hessians, pname): + def hook_fn(module, input, output): + x = output.detach().float() + if x.ndim == 3: + x = x.reshape(-1, x.shape[-1]) + hessians[pname] += (x.T @ x).cpu() + return hook_fn + +def collect_hessians(hessian_model, train_loader, args, device, grad_accum_steps, num_batches=256): + """Run calibration batches through a non-banked model, collecting H = X^T X for each CastedLinear. + Also captures the pre-logit hidden states for tok_emb.weight GPTQ (embedding GPTQ).""" + hessians = {} + hooks = [] + for name, module in hessian_model.named_modules(): + if isinstance(module, CastedLinear): + param_name = name + ".weight" + cols = module.weight.shape[1] + hessians[param_name] = torch.zeros(cols, cols, dtype=torch.float32, device='cpu') + h = module.register_forward_hook(_make_input_hook(hessians, param_name)) + hooks.append(h) + # Embedding GPTQ: capture Hessian for tok_emb.weight via pre-logit hidden states. + # The tied embedding acts as a linear layer: logits = hidden @ tok_emb.weight.T + # so the activation Hessian is X^T X where X = final_norm output. + embed_hessian_key = "tok_emb.weight" + embed_dim = hessian_model.tok_emb.weight.shape[1] + hessians[embed_hessian_key] = torch.zeros(embed_dim, embed_dim, dtype=torch.float32, device='cpu') + embed_hook = hessian_model.final_norm.register_forward_hook( + _make_output_hook(hessians, embed_hessian_key) + ) + hooks.append(embed_hook) + hessian_model.eval() + with torch.inference_mode(), torch.autocast(device_type="cuda", dtype=torch.bfloat16): + for _ in range(num_batches): + x, y = train_loader.next_batch(args.train_batch_tokens, args.train_seq_len, grad_accum_steps) + hessian_model(x, y) + for h in hooks: + h.remove() + for name in hessians: + H = hessians[name] + H /= num_batches + damp = 0.01 * torch.diag(H).mean().clamp_min(1e-6) + H += damp * torch.eye(H.shape[0]) + hessians[name] = H + hessian_model.train() + return hessians + +def mixed_quantize_int6(state_dict: dict[str, Tensor], int6_cats: set[str], hessians: dict[str, Tensor] | None = None, + matrix_bits: int = 6, embed_bits: int = 8, + matrix_clip_sigmas: float = 12.85, embed_clip_sigmas: float = 20.0): + """Mixed-precision GPTQ quantization. + - mlp/attn weights: int6 with SDClip (k=matrix_clip_sigmas) and Hessian compensation + - tok_emb/lm_head: GPTQ with int8 and SDClip (k=embed_clip_sigmas) + - small tensors: passthrough float16 + """ + num_layers_total = max( + (int(k.split(".")[1]) for k in state_dict if k.startswith("blocks.")), + default=0, + ) + 1 + result: dict[str, Tensor] = {} + meta: dict[str, object] = {} + for name, tensor in state_dict.items(): + t = tensor.detach().cpu().contiguous() + cat = _classify_param(name) + if not t.is_floating_point() or t.numel() <= 65536: + result[name] = t.to(torch.float16) if t.is_floating_point() else t + meta[name] = "passthrough" + continue + if any(p in name for p in CONTROL_TENSOR_NAME_PATTERNS): + result[name] = t.float() + meta[name] = "passthrough_ctrl" + continue + if cat == "embed" and t.ndim >= 1: + # GPTQ on embeddings with int8 + SDClip + cr = 2 ** (embed_bits - 1) - 1 # 127 for int8 + H = hessians.get(name) if hessians else None + if H is not None: + q, s = quantize_int6_gptq(t, hessian=H, clip_range=cr, clip_sigmas=embed_clip_sigmas) + else: + q, s = quantize_int6_per_row(t, clip_range=cr, clip_sigmas=embed_clip_sigmas) + result[name + ".q"] = q + result[name + ".scale"] = s + meta[name] = {"type": f"int{embed_bits}"} + elif cat in int6_cats and t.ndim >= 1: + cr = 2 ** (matrix_bits - 1) - 1 # 31 for int6 + H = hessians.get(name) if hessians else None + if H is not None: + q, s = quantize_int6_gptq(t, hessian=H, clip_range=cr, clip_sigmas=matrix_clip_sigmas) + else: + q, s = quantize_int6_per_row(t, clip_range=cr, clip_sigmas=matrix_clip_sigmas) + result[name + ".q"] = q + result[name + ".scale"] = s + meta[name] = {"type": f"int{matrix_bits}"} + else: + q, s = quantize_float_tensor(t) + result[name + ".q"] = q + result[name + ".scale"] = s + meta[name] = {"type": "int8"} + return result, meta +def dequantize_mixed_int6(result: dict[str, Tensor], meta: dict[str, object], + template_sd: dict[str, Tensor]) -> dict[str, Tensor]: + out: dict[str, Tensor] = {} + for name, orig in template_sd.items(): + info = meta.get(name) + if info is None: + continue + orig_dtype = orig.dtype + if info in ("passthrough", "passthrough_ctrl", "passthrough_fp16"): + t = result[name] + if t.dtype == torch.float16 and orig_dtype in (torch.float32, torch.bfloat16): + t = t.to(orig_dtype) + out[name] = t + continue + q, s = result[name + ".q"], result[name + ".scale"] + if s.ndim > 0: + out[name] = (q.float() * s.float().view(q.shape[0], *([1] * (q.ndim - 1)))).to(orig_dtype) + else: + out[name] = (q.float() * float(s.item())).to(orig_dtype) + return out -# ----------------------------- -# TRAINING -# ----------------------------- +# --- Byte-shuffle + Brotli compression --- + +_BSHF_MAGIC = b'BSHF' + +def _byte_shuffle(data: bytes, stride: int = 2) -> bytes: + """Interleave bytes by stride before compression (improves Brotli ratio on quantized weights).""" + if stride <= 1 or len(data) < stride: + return data + src = np.frombuffer(data, dtype=np.uint8) + n = len(src) + out = np.empty(n, dtype=np.uint8) + dest_off = 0 + for pos in range(stride): + chunk = src[pos::stride] + out[dest_off:dest_off + len(chunk)] = chunk + dest_off += len(chunk) + return _BSHF_MAGIC + bytes([stride]) + out.tobytes() + +def _byte_unshuffle(data: bytes) -> bytes: + """Reverse byte shuffle.""" + if len(data) < 5 or data[:4] != _BSHF_MAGIC: + return data + stride = data[4] + if stride < 2: + return data[5:] + payload = np.frombuffer(data, dtype=np.uint8, offset=5) + n = len(payload) + out = np.empty(n, dtype=np.uint8) + src_off = 0 + for pos in range(stride): + chunk_len = n // stride + (1 if pos < n % stride else 0) + out[pos::stride][:chunk_len] = payload[src_off:src_off + chunk_len] + src_off += chunk_len + return out.tobytes() + +def _compress_brotli(data: bytes) -> bytes: + """Byte-shuffle then Brotli compress.""" + import brotli + shuffled = _byte_shuffle(data, stride=2) + return brotli.compress(shuffled, quality=11) + +def _decompress_brotli(data: bytes) -> bytes: + """Brotli decompress then byte-unshuffle.""" + import brotli + raw = brotli.decompress(data) + return _byte_unshuffle(raw) + +# --- Training --- def main() -> None: - global zeropower_via_newtonschulz5 - code = Path(__file__).read_text(encoding="utf-8") args = Hyperparameters() - zeropower_via_newtonschulz5 = torch.compile(zeropower_via_newtonschulz5) - - # ----------------------------- - # DISTRIBUTED + CUDA SETUP - # ----------------------------- - + # zeropower_via_newtonschulz5 runs eagerly with bmm -- do NOT compile distributed = "RANK" in os.environ and "WORLD_SIZE" in os.environ rank = int(os.environ.get("RANK", "0")) world_size = int(os.environ.get("WORLD_SIZE", "1")) @@ -757,23 +1740,18 @@ def main() -> None: dist.init_process_group(backend="nccl", device_id=device) dist.barrier() master_process = rank == 0 - - # Fast math knobs torch.backends.cuda.matmul.allow_tf32 = True torch.backends.cudnn.allow_tf32 = True from torch.backends.cuda import enable_cudnn_sdp, enable_flash_sdp, enable_math_sdp, enable_mem_efficient_sdp - enable_cudnn_sdp(False) enable_flash_sdp(True) enable_mem_efficient_sdp(False) enable_math_sdp(False) - logfile = None if master_process: os.makedirs("logs", exist_ok=True) logfile = f"logs/{args.run_id}.txt" print(logfile) - def log0(msg: str, console: bool = True) -> None: if not master_process: return @@ -782,7 +1760,6 @@ def log0(msg: str, console: bool = True) -> None: if logfile is not None: with open(logfile, "a", encoding="utf-8") as f: print(msg, file=f) - log0(code, console=False) log0("=" * 100, console=False) log0(f"Running Python {sys.version}", console=False) @@ -792,16 +1769,10 @@ def log0(msg: str, console: bool = True) -> None: console=False, ) log0("=" * 100, console=False) - - # ----------------------------- - # TOKENIZER + VALIDATION METRIC SETUP - # ----------------------------- - random.seed(args.seed) np.random.seed(args.seed) torch.manual_seed(args.seed) torch.cuda.manual_seed_all(args.seed) - if not args.tokenizer_path.endswith(".model"): raise ValueError(f"Script only setup for SentencePiece .model file: {args.tokenizer_path}") sp = spm.SentencePieceProcessor(model_file=args.tokenizer_path) @@ -811,18 +1782,16 @@ def log0(msg: str, console: bool = True) -> None: ) dataset_dir = Path(args.data_path).resolve() actual_train_files = len(list(dataset_dir.glob("fineweb_train_*.bin"))) - val_tokens = load_validation_tokens(args.val_files, args.train_seq_len) + effective_eval_seq_len = args.eval_seq_len if args.eval_seq_len > 0 else args.train_seq_len + val_seq_len = max(args.train_seq_len, effective_eval_seq_len) + val_tokens = load_validation_tokens(args.val_files, val_seq_len) base_bytes_lut, has_leading_space_lut, is_boundary_token_lut = build_sentencepiece_luts( sp, args.vocab_size, device ) log0(f"val_bpb:enabled tokenizer_kind=sentencepiece tokenizer_path={args.tokenizer_path}") log0(f"train_loader:dataset:{dataset_dir.name} train_shards:{actual_train_files}") log0(f"val_loader:shards pattern={args.val_files} tokens:{val_tokens.numel() - 1}") - - # ----------------------------- - # MODEL + OPTIMIZER SETUP - # ----------------------------- - + CastedLinear._qat_enabled = args.qat_enabled base_model = GPT( vocab_size=args.vocab_size, num_layers=args.num_layers, @@ -835,25 +1804,45 @@ def log0(msg: str, console: bool = True) -> None: logit_softcap=args.logit_softcap, rope_base=args.rope_base, qk_gain_init=args.qk_gain_init, + mtp_num_heads=args.mtp_num_heads, + mtp_loss_weight=args.mtp_loss_weight, + bigram_vocab_size=args.bigram_vocab_size, + bigram_dim=args.bigram_dim, + xsa_last_n=args.xsa_last_n, + rope_dims=args.rope_dims, + ln_scale=args.ln_scale, + dtg=args.dtg_enabled, + ve_enabled=args.ve_enabled, + ve_dim=args.ve_dim, + ve_layers=args.ve_layers, + gated_attention=args.gated_attention, + value_residual=args.value_residual, + skip_gates_enabled=args.skip_gates_enabled, ).to(device).bfloat16() + # Banks stay FP32 (like CastedLinear weights), cast to BF16 in forward + base_model.qo_bank.data = base_model.qo_bank.data.float() + base_model.kv_bank.data = base_model.kv_bank.data.float() + base_model.mlp_up_bank.data = base_model.mlp_up_bank.data.float() + base_model.mlp_down_bank.data = base_model.mlp_down_bank.data.float() for module in base_model.modules(): if isinstance(module, CastedLinear): module.float() restore_low_dim_params_to_fp32(base_model) + # No DDP -- Parallel Muon handles bank grad communication via reduce-scatter, + # and non-bank grads are manually all-reduced before Adam steps. compiled_model = torch.compile(base_model, dynamic=False, fullgraph=True) - model: nn.Module = DDP(compiled_model, device_ids=[local_rank], broadcast_buffers=False) if distributed else compiled_model + model = compiled_model # Optimizer split: - # - token embedding (Adam) uses EMBED_LR - # - untied lm_head (Adam) uses HEAD_LR - # - matrix params in transformer blocks use MATRIX_LR via Muon - # - vectors/scalars use SCALAR_LR via Adam - block_named_params = list(base_model.blocks.named_parameters()) + # - 4 parameter banks -> Muon (batched Newton-Schulz) + # - token embedding -> Adam + # - scalars/control tensors -> Adam + # - bigram proj, mtp heads, VE proj -> Adam (small matrix params not worth banking) matrix_params = [ - p - for name, p in block_named_params - if p.ndim == 2 and not any(pattern in name for pattern in CONTROL_TENSOR_NAME_PATTERNS) + base_model.qo_bank, base_model.kv_bank, + base_model.mlp_up_bank, base_model.mlp_down_bank, ] + block_named_params = list(base_model.blocks.named_parameters()) scalar_params = [ p for name, p in block_named_params @@ -861,11 +1850,29 @@ def log0(msg: str, console: bool = True) -> None: ] if base_model.skip_weights.numel() > 0: scalar_params.append(base_model.skip_weights) + if base_model.skip_gates is not None and base_model.skip_gates.numel() > 0: + scalar_params.append(base_model.skip_gates) + scalar_params.append(base_model.smear.gate) + if base_model.bigram is not None: + scalar_params.append(base_model.bigram.scale) token_lr = args.tied_embed_lr if args.tie_embeddings else args.embed_lr - optimizer_tok = torch.optim.Adam( - [{"params": [base_model.tok_emb.weight], "lr": token_lr, "base_lr": token_lr}], + tok_params = [{"params": [base_model.tok_emb.weight], "lr": token_lr, "base_lr": token_lr}] + if base_model.bigram is not None: + tok_params.append({"params": [base_model.bigram.embed.weight], "lr": token_lr, "base_lr": token_lr}) + if base_model.bigram.proj is not None: + scalar_params.append(base_model.bigram.proj.weight) + if base_model.ve_shared is not None: + tok_params.append({"params": [base_model.ve_shared.embed.weight], "lr": token_lr, "base_lr": token_lr}) + if base_model.ve_shared.proj is not None: + scalar_params.append(base_model.ve_shared.proj.weight) + scalar_params.append(base_model.ve_shared.scale) + for s in base_model.ve_layer_scales: + scalar_params.append(s) + optimizer_tok = torch.optim.AdamW( + tok_params, betas=(args.beta1, args.beta2), eps=args.adam_eps, + weight_decay=args.embed_wd, fused=True, ) optimizer_muon = Muon( @@ -873,16 +1880,24 @@ def log0(msg: str, console: bool = True) -> None: lr=args.matrix_lr, momentum=args.muon_momentum, backend_steps=args.muon_backend_steps, + weight_decay=args.muon_wd, ) for group in optimizer_muon.param_groups: group["base_lr"] = args.matrix_lr - optimizer_scalar = torch.optim.Adam( + optimizer_scalar = torch.optim.AdamW( [{"params": scalar_params, "lr": args.scalar_lr, "base_lr": args.scalar_lr}], betas=(args.beta1, args.beta2), eps=args.adam_eps, + weight_decay=args.adam_wd, fused=True, ) - optimizers: list[torch.optim.Optimizer] = [optimizer_tok, optimizer_muon, optimizer_scalar] + # Non-bank params that need manual all-reduce (replicated across GPUs) + replicated_params = list(optimizer_tok.param_groups[0]["params"]) + for pg in optimizer_tok.param_groups[1:]: + replicated_params.extend(pg["params"]) + replicated_params.extend(scalar_params) + + optimizer_head = None if base_model.lm_head is not None: optimizer_head = torch.optim.Adam( [{"params": [base_model.lm_head.weight], "lr": args.head_lr, "base_lr": args.head_lr}], @@ -890,10 +1905,16 @@ def log0(msg: str, console: bool = True) -> None: eps=args.adam_eps, fused=True, ) - optimizers.insert(1, optimizer_head) - + replicated_params.append(base_model.lm_head.weight) + optimizers: list[torch.optim.Optimizer] = [optimizer_tok, optimizer_muon, optimizer_scalar] + if optimizer_head is not None: + optimizers.append(optimizer_head) n_params = sum(p.numel() for p in base_model.parameters()) + mtp_params = sum(p.numel() for p in base_model.mtp_heads.parameters()) log0(f"model_params:{n_params}") + log0(f"mtp_num_heads:{args.mtp_num_heads} mtp_loss_weight:{args.mtp_loss_weight} mtp_params:{mtp_params}") + xsa_layers = [i for i, b in enumerate(base_model.blocks) if b.attn.use_xsa] + log0(f"XSA:last_{args.xsa_last_n} active_layers:{xsa_layers}") log0(f"world_size:{world_size} grad_accum_steps:{grad_accum_steps}") log0("sdp_backends:cudnn=False flash=True mem_efficient=False math=False") log0(f"attention_mode:gqa num_heads:{args.num_heads} num_kv_heads:{args.num_kv_heads}") @@ -908,20 +1929,22 @@ def log0(msg: str, console: bool = True) -> None: f"max_wallclock_seconds:{args.max_wallclock_seconds:.3f}" ) log0(f"seed:{args.seed}") - - # ----------------------------- - # DATA LOADER & MODEL WARMUP - # ----------------------------- - train_loader = DistributedTokenLoader(args.train_files, rank, world_size, device) - def zero_grad_all() -> None: for opt in optimizers: opt.zero_grad(set_to_none=True) - max_wallclock_ms = 1000.0 * args.max_wallclock_seconds if args.max_wallclock_seconds > 0 else None - + def training_frac(step: int, elapsed_ms: float) -> float: + if max_wallclock_ms is None: + return step / max(args.iterations, 1) + return elapsed_ms / max(max_wallclock_ms, 1e-9) def lr_mul(step: int, elapsed_ms: float) -> float: + # warmdown_frac > 0 takes priority; fallback to legacy warmdown_iters + if args.warmdown_frac > 0: + frac = training_frac(step, elapsed_ms) + if frac >= 1.0 - args.warmdown_frac: + return max((1.0 - frac) / args.warmdown_frac, 0.0) + return 1.0 if args.warmdown_iters <= 0: return 1.0 if max_wallclock_ms is None: @@ -931,9 +1954,6 @@ def lr_mul(step: int, elapsed_ms: float) -> float: warmdown_ms = args.warmdown_iters * step_ms remaining_ms = max(max_wallclock_ms - elapsed_ms, 0.0) return remaining_ms / max(warmdown_ms, 1e-9) if remaining_ms <= warmdown_ms else 1.0 - - # Warmup primes the compiled forward/backward/optimizer paths, then we restore the - # initial weights/optimizer state so measured training starts from the true init. if args.warmup_steps > 0: initial_model_state = {name: tensor.detach().cpu().clone() for name, tensor in base_model.state_dict().items()} initial_optimizer_states = [copy.deepcopy(opt.state_dict()) for opt in optimizers] @@ -941,12 +1961,15 @@ def lr_mul(step: int, elapsed_ms: float) -> float: for warmup_step in range(args.warmup_steps): zero_grad_all() for micro_step in range(grad_accum_steps): - if distributed: - model.require_backward_grad_sync = micro_step == grad_accum_steps - 1 x, y = train_loader.next_batch(args.train_batch_tokens, args.train_seq_len, grad_accum_steps) with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): warmup_loss = model(x, y) (warmup_loss * grad_scale).backward() + # All-reduce all grads for warmup (simple, not optimized) + if distributed: + for p in base_model.parameters(): + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG) for opt in optimizers: opt.step() zero_grad_all() @@ -956,23 +1979,20 @@ def lr_mul(step: int, elapsed_ms: float) -> float: for opt, state in zip(optimizers, initial_optimizer_states, strict=True): opt.load_state_dict(state) zero_grad_all() - if distributed: - model.require_backward_grad_sync = True train_loader = DistributedTokenLoader(args.train_files, rank, world_size, device) - - # ----------------------------- - # MAIN TRAINING LOOP - # ----------------------------- - + swa_state: dict[str, Tensor] | None = None + swa_count = 0 + from collections import deque + lawa_queue: deque[dict[str, Tensor]] = deque(maxlen=args.lawa_k) + ema_state = {name: t.detach().float().clone() for name, t in base_model.state_dict().items()} + ema_decay = 0.997 training_time_ms = 0.0 stop_after_step: int | None = None torch.cuda.synchronize() t0 = time.perf_counter() - step = 0 while True: last_step = step == args.iterations or (stop_after_step is not None and step >= stop_after_step) - should_validate = last_step or (args.val_loss_every > 0 and step % args.val_loss_every == 0) if should_validate: torch.cuda.synchronize() @@ -995,7 +2015,6 @@ def lr_mul(step: int, elapsed_ms: float) -> float: ) torch.cuda.synchronize() t0 = time.perf_counter() - if last_step: if stop_after_step is not None and step < args.iterations: log0( @@ -1003,38 +2022,61 @@ def lr_mul(step: int, elapsed_ms: float) -> float: f"step:{step}/{args.iterations}" ) break - elapsed_ms = training_time_ms + 1000.0 * (time.perf_counter() - t0) scale = lr_mul(step, elapsed_ms) + if args.late_qat_threshold > 0 and scale < args.late_qat_threshold and not CastedLinear._qat_enabled: + CastedLinear._qat_enabled = True + log0(f"late_qat:enabled step:{step} scale:{scale:.4f}") zero_grad_all() train_loss = torch.zeros((), device=device) for micro_step in range(grad_accum_steps): - if distributed: - model.require_backward_grad_sync = micro_step == grad_accum_steps - 1 x, y = train_loader.next_batch(args.train_batch_tokens, args.train_seq_len, grad_accum_steps) with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): loss = model(x, y) train_loss += loss.detach() (loss * grad_scale).backward() train_loss /= grad_accum_steps - frac = min(step / args.muon_momentum_warmup_steps, 1.0) if args.muon_momentum_warmup_steps > 0 else 1.0 muon_momentum = (1 - frac) * args.muon_momentum_warmup_start + frac * args.muon_momentum for group in optimizer_muon.param_groups: group["momentum"] = muon_momentum - for opt in optimizers: for group in opt.param_groups: group["lr"] = group["base_lr"] * scale - if args.grad_clip_norm > 0: torch.nn.utils.clip_grad_norm_(base_model.parameters(), args.grad_clip_norm) - for opt in optimizers: - opt.step() + # === 3-phase overlapped optimizer step === + # Phase 1: Launch async reduce-scatter for banks (biggest first) + optimizer_muon.launch_reduce_scatters() + # Phase 2: All-reduce non-bank grads + step Adam (while bank RS is in-flight) + if distributed: + for p in replicated_params: + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG) + optimizer_tok.step() + optimizer_scalar.step() + if optimizer_head is not None: + optimizer_head.step() + # Phase 3: Wait for RS, local NS5, all-gather (banks processed last) + optimizer_muon.step() zero_grad_all() - + # EMA update + with torch.no_grad(): + for name, t in base_model.state_dict().items(): + ema_state[name].mul_(ema_decay).add_(t.detach().float(), alpha=1.0 - ema_decay) step += 1 approx_training_time_ms = training_time_ms + 1000.0 * (time.perf_counter() - t0) + if args.swa_enabled and scale < 0.2 and step % args.swa_every == 0: + if swa_state is None: + swa_state = {name: t.detach().cpu().clone() for name, t in base_model.state_dict().items()} + swa_count = 1 + log0(f"swa:start step:{step}") + else: + for name, t in base_model.state_dict().items(): + swa_state[name] += t.detach().cpu() + swa_count += 1 + if args.lawa_enabled and step % args.lawa_freq == 0: + lawa_queue.append({name: t.detach().cpu().clone() for name, t in base_model.state_dict().items()}) should_log_train = ( args.train_log_every > 0 and (step <= 10 or step % args.train_log_every == 0 or stop_after_step is not None) @@ -1044,8 +2086,6 @@ def lr_mul(step: int, elapsed_ms: float) -> float: f"step:{step}/{args.iterations} train_loss:{train_loss.item():.4f} " f"train_time:{approx_training_time_ms:.0f}ms step_avg:{approx_training_time_ms / step:.2f}ms" ) - - # Needed to sync whether we've reached the wallclock cap. reached_cap = max_wallclock_ms is not None and approx_training_time_ms >= max_wallclock_ms if distributed and max_wallclock_ms is not None: reached_cap_tensor = torch.tensor(int(reached_cap), device=device) @@ -1053,74 +2093,326 @@ def lr_mul(step: int, elapsed_ms: float) -> float: reached_cap = bool(reached_cap_tensor.item()) if stop_after_step is None and reached_cap: stop_after_step = step - log0( f"peak memory allocated: {torch.cuda.max_memory_allocated() // 1024 // 1024} MiB " f"reserved: {torch.cuda.max_memory_reserved() // 1024 // 1024} MiB" ) - - # ----------------------------- - # SERIALIZATION + ROUNDTRIP VALIDATION - # ----------------------------- - # Save the raw state (useful for debugging/loading in PyTorch directly), then always produce - # the compressed int8+zlib artifact and validate the round-tripped weights. - + # Apply weight averaging + if args.lawa_enabled and len(lawa_queue) > 1: + log0(f"lawa:applying LAWA averaging k={len(lawa_queue)}") + current_state = base_model.state_dict() + avg_state = {name: torch.zeros(t.shape, dtype=torch.float32, device='cpu') for name, t in current_state.items()} + for snap in lawa_queue: + for name in avg_state: + avg_state[name] += snap[name].float() + for name in avg_state: + avg_state[name] /= len(lawa_queue) + avg_state[name] = avg_state[name].to(dtype=current_state[name].dtype) + base_model.load_state_dict(avg_state, strict=True) + else: + log0("ema:applying EMA weights") + current_state = base_model.state_dict() + avg_state = {name: t.to(dtype=current_state[name].dtype) for name, t in ema_state.items()} + base_model.load_state_dict(avg_state, strict=True) + torch.cuda.synchronize() + t_diag = time.perf_counter() + diag_val_loss, diag_val_bpb = eval_val( + args, compiled_model, rank, world_size, device, grad_accum_steps, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + ) + torch.cuda.synchronize() + log0( + f"DIAGNOSTIC post_ema val_loss:{diag_val_loss:.4f} val_bpb:{diag_val_bpb:.4f} " + f"eval_time:{1000.0 * (time.perf_counter() - t_diag):.0f}ms" + ) + # AdamW TTT: fine-tune EMA model on val data BEFORE quantization + if args.ttt_enabled: + if distributed: + dist.barrier() + log0(f"ttt:start lr={args.ttt_lr} epochs={args.ttt_epochs} " + f"freeze_blocks={args.ttt_freeze_blocks} cosine_decay={args.ttt_cosine_decay}") + t_ttt = time.perf_counter() + ttt_adapt_adamw( + args, base_model, device, val_tokens, + rank=rank, world_size=world_size, log0=log0, + ) + torch.cuda.synchronize() + log0(f"ttt:elapsed={time.perf_counter() - t_ttt:.1f}s") + t_ttt_diag = time.perf_counter() + ttt_diag_loss, ttt_diag_bpb = eval_val( + args, base_model, rank, world_size, device, grad_accum_steps, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + ) + torch.cuda.synchronize() + log0(f"DIAGNOSTIC post_ttt val_loss:{ttt_diag_loss:.4f} val_bpb:{ttt_diag_bpb:.4f} " + f"eval_time:{1000.0 * (time.perf_counter() - t_ttt_diag):.0f}ms") + if distributed: + dist.barrier() + full_state_dict = base_model.state_dict() + export_sd = {k: v for k, v in full_state_dict.items() if "mtp_heads" not in k} + excluded_mtp = sum(int(t.numel()) for k, t in full_state_dict.items() if "mtp_heads" in k) + if excluded_mtp > 0: + log0(f"export_excluding_mtp_params:{excluded_mtp}") if master_process: - torch.save(base_model.state_dict(), "final_model.pt") + torch.save(export_sd, "final_model.pt") model_bytes = os.path.getsize("final_model.pt") code_bytes = len(code.encode("utf-8")) log0(f"Serialized model: {model_bytes} bytes") log0(f"Code size: {code_bytes} bytes") - log0(f"Total submission size: {model_bytes + code_bytes} bytes") - - quant_obj, quant_stats = quantize_state_dict_int8(base_model.state_dict()) + # Unbank 3D tensors into individual 2D tensors for quantization + sd_cpu = {k: v.detach().cpu() for k, v in export_sd.items()} + unbanked_sd = _unbank_state_dict(sd_cpu, args.num_layers) + # Full GPTQ: collect Hessians via a temporary non-banked model + log0(f"gptq:building non-banked model for Hessian collection...") + hessian_model = _HessianGPT( + vocab_size=args.vocab_size, num_layers=args.num_layers, model_dim=args.model_dim, + num_heads=args.num_heads, num_kv_heads=args.num_kv_heads, mlp_mult=args.mlp_mult, + tie_embeddings=args.tie_embeddings, logit_softcap=args.logit_softcap, + rope_base=args.rope_base, qk_gain_init=args.qk_gain_init, + bigram_vocab_size=args.bigram_vocab_size, bigram_dim=args.bigram_dim, + xsa_last_n=args.xsa_last_n, rope_dims=args.rope_dims, ln_scale=args.ln_scale, + ve_enabled=args.ve_enabled, ve_dim=args.ve_dim, ve_layers=args.ve_layers, + ).to(device).bfloat16() + for m in hessian_model.modules(): + if isinstance(m, CastedLinear): + m.float() + restore_low_dim_params_to_fp32(hessian_model) + # Load unbanked weights into the non-banked model + hessian_model.load_state_dict( + {k: v.to(device) for k, v in unbanked_sd.items() if k in hessian_model.state_dict()}, + strict=False, + ) + # Training-data calibration (proven -0.0007 BPP vs AR self-gen) + log0(f"gptq:collecting hessians from training data ({args.gptq_calib_batches} batches)...") + hessians = collect_hessians(hessian_model, train_loader, args, device, grad_accum_steps, num_batches=args.gptq_calib_batches) + log0(f"gptq:collected hessians for {len(hessians)} layers (training data)") + del hessian_model + torch.cuda.empty_cache() + quant_result, quant_meta = mixed_quantize_int6( + unbanked_sd, {"mlp", "attn"}, hessians=hessians, + matrix_bits=args.matrix_bits, embed_bits=args.embed_bits, + matrix_clip_sigmas=args.matrix_clip_sigmas, embed_clip_sigmas=args.embed_clip_sigmas, + ) + # NOVEL: Selective ±1 pruning by reconstruction error + # Sort ±1 quantized values by their reconstruction error (scale²), + # prune least-impactful first until artifact fits target size. + target_mb = float(os.environ.get("TARGET_MB", "15.9")) + code_bytes_est = len(code.encode("utf-8")) + ones_info = [] # (tensor_key, flat_idx, error) + for name, info in quant_meta.items(): + if not (isinstance(info, dict) and info.get("type") == "int6"): continue + qk, sk = name + ".q", name + ".scale" + if qk not in quant_result or sk not in quant_result: continue + q, s = quant_result[qk], quant_result[sk] + if s.ndim > 0: + ones_mask = (q.abs() == 1) + if ones_mask.any(): + row_idx = torch.arange(q.shape[0]).unsqueeze(1).expand_as(q)[ones_mask] + flat_idx = torch.arange(q.numel()).reshape(q.shape)[ones_mask] + errors = s.float()[row_idx].pow(2) + for fi, err in zip(flat_idx.tolist(), errors.tolist()): + ones_info.append((qk, fi, err)) + if ones_info: + ones_info.sort(key=lambda x: x[2]) + def _try_prune(n): + tmp = {k: v.clone() for k, v in quant_result.items()} + for i in range(min(n, len(ones_info))): + tmp[ones_info[i][0]].view(-1)[ones_info[i][1]] = 0 + buf = io.BytesIO(); torch.save({"w": tmp, "m": quant_meta}, buf) + return len(_compress_brotli(buf.getvalue())) + code_bytes_est, tmp + no_sz, _ = _try_prune(0) + target_bytes = int(target_mb * 1024 * 1024) + log0(f"selective_prune: {len(ones_info)} ±1 candidates, unpruned={no_sz/(1024*1024):.2f}MB target={target_mb}MB") + if no_sz <= target_bytes: + log0("selective_prune: already fits, no pruning needed") + else: + full_sz, _ = _try_prune(len(ones_info)) + log0(f"selective_prune: full ±1 prune={full_sz/(1024*1024):.2f}MB") + if full_sz > target_bytes: + log0("selective_prune: even full prune not enough, applying all") + _, quant_result = _try_prune(len(ones_info)) + else: + lo, hi = 0, len(ones_info) + while lo < hi: + mid = (lo + hi) // 2 + sz, _ = _try_prune(mid) + if sz <= target_bytes: hi = mid + else: lo = mid + 1 + log0(f"selective_prune: pruning {lo}/{len(ones_info)} ±1 values ({100*lo/len(ones_info):.1f}%) to fit {target_mb}MB") + _, quant_result = _try_prune(lo) quant_buf = io.BytesIO() - torch.save(quant_obj, quant_buf) + torch.save({"w": quant_result, "m": quant_meta}, quant_buf) quant_raw = quant_buf.getvalue() - quant_blob = zlib.compress(quant_raw, level=9) - quant_raw_bytes = len(quant_raw) + quant_blob = _compress_brotli(quant_raw) if master_process: - with open("final_model.int8.ptz", "wb") as f: + with open("final_model.int6.ptz", "wb") as f: f.write(quant_blob) - quant_file_bytes = os.path.getsize("final_model.int8.ptz") + quant_file_bytes = len(quant_blob) code_bytes = len(code.encode("utf-8")) - ratio = quant_stats["baseline_tensor_bytes"] / max(quant_stats["int8_payload_bytes"], 1) - log0( - f"Serialized model int8+zlib: {quant_file_bytes} bytes " - f"(payload:{quant_stats['int8_payload_bytes']} raw_torch:{quant_raw_bytes} payload_ratio:{ratio:.2f}x)" - ) - log0(f"Total submission size int8+zlib: {quant_file_bytes + code_bytes} bytes") - + log0(f"Serialized model int6+brotli+byteshuffle: {quant_file_bytes} bytes") + log0(f"Total submission size int6+brotli+byteshuffle: {quant_file_bytes + code_bytes} bytes") if distributed: dist.barrier() - with open("final_model.int8.ptz", "rb") as f: + with open("final_model.int6.ptz", "rb") as f: quant_blob_disk = f.read() - quant_state = torch.load(io.BytesIO(zlib.decompress(quant_blob_disk)), map_location="cpu") - base_model.load_state_dict(dequantize_state_dict_int8(quant_state), strict=True) + quant_state = torch.load( + io.BytesIO(_decompress_brotli(quant_blob_disk)), + map_location="cpu", + weights_only=False, + ) + deq_unbanked = dequantize_mixed_int6(quant_state["w"], quant_state["m"], unbanked_sd) + # Re-bank the dequantized tensors + deq_state = _rebank_state_dict(deq_unbanked, args.num_layers, sd_cpu) + eval_model = GPT( + vocab_size=args.vocab_size, num_layers=args.num_layers, model_dim=args.model_dim, + num_heads=args.num_heads, num_kv_heads=args.num_kv_heads, mlp_mult=args.mlp_mult, + tie_embeddings=args.tie_embeddings, tied_embed_init_std=args.tied_embed_init_std, + logit_softcap=args.logit_softcap, rope_base=args.rope_base, qk_gain_init=args.qk_gain_init, + mtp_num_heads=0, mtp_loss_weight=0.0, + bigram_vocab_size=args.bigram_vocab_size, bigram_dim=args.bigram_dim, + xsa_last_n=args.xsa_last_n, + rope_dims=args.rope_dims, ln_scale=args.ln_scale, dtg=args.dtg_enabled, + ve_enabled=args.ve_enabled, ve_dim=args.ve_dim, ve_layers=args.ve_layers, + gated_attention=args.gated_attention, value_residual=args.value_residual, + skip_gates_enabled=args.skip_gates_enabled, + ).to(device).bfloat16() + eval_model.qo_bank.data = eval_model.qo_bank.data.float() + eval_model.kv_bank.data = eval_model.kv_bank.data.float() + eval_model.mlp_up_bank.data = eval_model.mlp_up_bank.data.float() + eval_model.mlp_down_bank.data = eval_model.mlp_down_bank.data.float() + for m in eval_model.modules(): + if isinstance(m, CastedLinear): + m.float() + restore_low_dim_params_to_fp32(eval_model) + eval_model.load_state_dict(deq_state, strict=True) + compiled_eval = torch.compile(eval_model, dynamic=False, fullgraph=True) torch.cuda.synchronize() t_qeval = time.perf_counter() q_val_loss, q_val_bpb = eval_val( - args, - model, - rank, - world_size, - device, - grad_accum_steps, - val_tokens, - base_bytes_lut, - has_leading_space_lut, - is_boundary_token_lut, + args, compiled_eval, rank, world_size, device, grad_accum_steps, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + eval_seq_len=effective_eval_seq_len, ) torch.cuda.synchronize() log0( - f"final_int8_zlib_roundtrip val_loss:{q_val_loss:.4f} val_bpb:{q_val_bpb:.4f} " + f"final_int6_roundtrip val_loss:{q_val_loss:.4f} val_bpb:{q_val_bpb:.4f} " f"eval_time:{1000.0 * (time.perf_counter() - t_qeval):.0f}ms" ) - log0(f"final_int8_zlib_roundtrip_exact val_loss:{q_val_loss:.8f} val_bpb:{q_val_bpb:.8f}") - + log0(f"final_int6_roundtrip_exact val_loss:{q_val_loss:.8f} val_bpb:{q_val_bpb:.8f}") + sw_seq_len = effective_eval_seq_len + if args.eval_stride > 0 and args.eval_stride < sw_seq_len: + torch.cuda.synchronize() + t_slide = time.perf_counter() + sw_val_loss, sw_val_bpb = eval_val_sliding( + args, eval_model, rank, world_size, device, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + stride=args.eval_stride, + eval_seq_len=sw_seq_len, + ) + torch.cuda.synchronize() + log0( + f"final_int6_sliding_window val_loss:{sw_val_loss:.4f} val_bpb:{sw_val_bpb:.4f} " + f"stride:{args.eval_stride} eval_time:{1000.0 * (time.perf_counter() - t_slide):.0f}ms" + ) + log0(f"final_int6_sliding_window_exact val_loss:{sw_val_loss:.8f} val_bpb:{sw_val_bpb:.8f}") + log0(f"final_int8_zlib_roundtrip_exact val_loss:{sw_val_loss:.8f} val_bpb:{sw_val_bpb:.8f}") + if args.eval_stride != 64 and 64 < sw_seq_len: + torch.cuda.synchronize() + t_slide64 = time.perf_counter() + sw64_val_loss, sw64_val_bpb = eval_val_sliding( + args, eval_model, rank, world_size, device, + val_tokens, base_bytes_lut, has_leading_space_lut, is_boundary_token_lut, + stride=64, + eval_seq_len=sw_seq_len, + ) + torch.cuda.synchronize() + log0( + f"final_int6_sliding_window_s64 val_loss:{sw64_val_loss:.4f} val_bpb:{sw64_val_bpb:.4f} " + f"stride:64 eval_time:{1000.0 * (time.perf_counter() - t_slide64):.0f}ms" + ) + log0(f"final_int6_sliding_window_s64_exact val_loss:{sw64_val_loss:.8f} val_bpb:{sw64_val_bpb:.8f}") + log0(f"final_int8_zlib_roundtrip_exact val_loss:{sw64_val_loss:.8f} val_bpb:{sw64_val_bpb:.8f}") + # --- Causal SLOT: per-batch delta optimization using ONLY context (already-scored) positions --- + if args.slot_enabled and args.eval_stride > 0 and args.eval_stride < sw_seq_len: + try: + slot_stride = args.eval_stride + seq_s = sw_seq_len + total_tok = val_tokens.numel() - 1 + ws_list = [ws for ws in range(0, total_tok, slot_stride) if min(ws + seq_s, total_tok) - ws >= 1] + my_s = (len(ws_list) * rank) // world_size + my_e = (len(ws_list) * (rank + 1)) // world_size + my_ws = ws_list[my_s:my_e] + num_batches = (len(my_ws) + 31) // 32 + sl_loss = torch.zeros((), device=device, dtype=torch.float64) + sl_tc = torch.zeros((), device=device, dtype=torch.float64) + sl_bc = torch.zeros((), device=device, dtype=torch.float64) + torch.cuda.synchronize() + t_slot = time.perf_counter() + eval_model.eval() + log0(f"causal_slot:start lr={args.slot_lr} steps={args.slot_steps} stride={slot_stride} " + f"windows={len(my_ws)} batches={num_batches}") + for batch_idx, bi in enumerate(range(0, len(my_ws), 32)): + bws = my_ws[bi:bi+32]; bsz = len(bws) + xb = torch.zeros(bsz, seq_s, dtype=torch.int64, device=device) + yb = torch.zeros(bsz, seq_s, dtype=torch.int64, device=device) + wls = [] + for i, ws in enumerate(bws): + end = min(ws + seq_s, total_tok); wl = end - ws; wls.append(wl) + ct = val_tokens[ws:end+1].to(dtype=torch.int64, device=device) + xb[i,:wl] = ct[:-1]; yb[i,:wl] = ct[1:] + # Frozen forward: get hidden states + with torch.no_grad(), torch.autocast(device_type="cuda", dtype=torch.bfloat16): + H = eval_model.forward_hidden(xb) + H = H.detach().float() + # Build context mask (already-scored positions) and score mask (new positions) + context_mask = torch.zeros(bsz, seq_s, dtype=torch.bool, device=device) + has_context = False + for i, ws in enumerate(bws): + wl = wls[i]; s = 0 if ws == 0 else max(wl - slot_stride, 0) + if s > 0: + context_mask[i, :s] = True + has_context = True + # Optimize delta on context positions only (CAUSAL: no future info) + delta = torch.zeros(1, 1, H.shape[-1], device=device, dtype=H.dtype, requires_grad=True) + if has_context: + sopt = torch.optim.AdamW([delta], lr=args.slot_lr, betas=(0.3, 0.9), weight_decay=1e-8, eps=1e-5) + for _ in range(args.slot_steps): + sopt.zero_grad() + lg = eval_model.compute_logits((H + delta).to(torch.bfloat16)).float() + nll_all = F.cross_entropy(lg.reshape(-1, lg.size(-1)), yb.reshape(-1), reduction="none").reshape(bsz, seq_s) + ctx_nll = nll_all[context_mask] + if ctx_nll.numel() > 0: + loss_c = ctx_nll.mean() + loss_c.backward() + sopt.step() + # Score new positions with adapted delta + with torch.no_grad(): + lg = eval_model.compute_logits((H + delta.detach()).to(torch.bfloat16)).float() + nll = F.cross_entropy(lg.reshape(-1, lg.size(-1)), yb.reshape(-1), reduction="none").reshape(bsz, seq_s) + for i, ws in enumerate(bws): + wl = wls[i]; s = 0 if ws == 0 else max(wl - slot_stride, 0) + sl_loss += nll[i, s:wl].to(torch.float64).sum(); sl_tc += float(wl - s) + tgt, prev = yb[i, s:wl], xb[i, s:wl] + tb = base_bytes_lut[tgt].to(torch.float64) + tb += (has_leading_space_lut[tgt] & ~is_boundary_token_lut[prev]).to(torch.float64) + sl_bc += tb.sum() + if batch_idx % 500 == 0 or batch_idx == num_batches - 1: + log0(f" causal_slot:batch {batch_idx+1}/{num_batches} " + f"time:{time.perf_counter()-t_slot:.1f}s"); sys.stdout.flush() + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(sl_loss, op=dist.ReduceOp.SUM) + dist.all_reduce(sl_tc, op=dist.ReduceOp.SUM) + dist.all_reduce(sl_bc, op=dist.ReduceOp.SUM) + sv_loss = (sl_loss / sl_tc).item() + sv_bpb = sv_loss / math.log(2.0) * (sl_tc.item() / sl_bc.item()) + torch.cuda.synchronize() + log0(f"final_causal_slot val_loss:{sv_loss:.4f} val_bpb:{sv_bpb:.4f} time:{1000*(time.perf_counter()-t_slot):.0f}ms") + log0(f"final_causal_slot_exact val_loss:{sv_loss:.8f} val_bpb:{sv_bpb:.8f}") + log0(f"final_int8_zlib_roundtrip_exact val_loss:{sv_loss:.8f} val_bpb:{sv_bpb:.8f}") + except Exception as e: + import traceback; log0(f"causal_slot:ERROR {e}"); traceback.print_exc(); sys.stdout.flush() if distributed: dist.destroy_process_group() - - + sys.exit(0) if __name__ == "__main__": main()