diff --git a/records/track_non_record_16mb/2026-04-01_ar_gptq_xsa_rotq_hadamard/README.md b/records/track_non_record_16mb/2026-04-01_ar_gptq_xsa_rotq_hadamard/README.md new file mode 100644 index 0000000000..4b1d5a4938 --- /dev/null +++ b/records/track_non_record_16mb/2026-04-01_ar_gptq_xsa_rotq_hadamard/README.md @@ -0,0 +1,136 @@ +# AR Self-Gen GPTQ + XSA-all + Per-Layer Hadamard ROTQ + +**Non-record submission.** + +This run uses the same evaluation metric and artifact accounting as the main leaderboard, but it is **not** a main-track record submission because it was trained on **1xH100 for 4800 seconds**, not reproduced under the official **8xH100 SXM / 600 second** budget. + +## Summary + +This submission starts from the current public `11L AR Self-Gen GPTQ + XSA` stack and adds a small, modular quantization-basis change: + +- keep the runtime model standard dense linear layers after load +- choose a compact right-rotation for large MLP matrices before GPTQ +- quantize in the rotated basis +- invert the rotation after dequantization during roundtrip evaluation + +The best variant here is simple: + +- Hadamard right-rotation +- applied to `mlp_up` and `mlp_down` +- block size chosen per layer from `{128, 256, 512}` + +On this longer single-GPU run, that export path slightly improves the exact final sliding score over the unrotated export on the same checkpoint while staying under the 16,000,000-byte cap. + +## Result + +Reference values from the current upstream README at submission time: + +- Main leaderboard target: `1.1147` +- Naive baseline: `1.2244` + +Best result in this folder: + +- Exact sliding `val_bpb`: `1.11290586` +- Exact sliding `val_loss`: `1.87908996` +- Total artifact bytes: `15,826,148` +- Hardware / duration: `1xH100`, `4800s`, seed `314` + +Because this is a non-record run, the correct interpretation is: + +- same metric as the leaderboard +- numerically better than the current top README score +- not eligible for the main track until reproduced within the official budget + +## Ablation Table + +All rows below use the same trained checkpoint and differ only in the export path. + +| Export path | Exact sliding val_bpb | Bytes total | +|---|---:|---:| +| Base AR self-gen GPTQ export | `1.11296252` | `15,869,601` | +| `mlp_down` Hadamard `256` | `1.11291713` | `15,825,000` | +| `mlp_down` per-layer Hadamard | `1.11290938` | `15,826,480` | +| `mlp_up + mlp_down` per-layer Hadamard | `1.11290586` | `15,826,148` | + +Net gain from the best rotation-aware export over the base export on the same checkpoint: + +- `-0.00005666` BPB + +## What Changed + +Relative to the public AR self-generated GPTQ + XSA-all baseline stack, this version adds: + +1. Rotation-aware GPTQ hooks for selected 2D tensors +2. Blockwise Walsh-Hadamard right-rotations before GPTQ +3. Shared or per-layer rotation selection over semantic tensor groups +4. Roundtrip reconstruction by inverting the right-rotation after dequantization + +The implementation stays modular and optional through environment flags. The best result here came from forcing per-layer Hadamard search on `mlp_up` and `mlp_down`. + +## Run Commands + +Training run used for the checkpoint: + +```bash +RUN_ID=ar_xsa_base_long4800_seed314 \ +DATA_PATH=/workspace/parameter-golf/data/datasets/fineweb10B_sp1024/ \ +TOKENIZER_PATH=/workspace/parameter-golf/data/tokenizers/fineweb_1024_bpe.model \ +VOCAB_SIZE=1024 \ +BIGRAM_VOCAB_SIZE=3072 \ +BIGRAM_DIM=112 \ +WARMDOWN_ITERS=4000 \ +TARGET_MB=15.9 \ +SEED=314 \ +MAX_WALLCLOCK_SECONDS=4800 \ +ROTQ_ENABLED=0 \ +EVAL_FORCE_S64=1 \ +torchrun --standalone --nproc_per_node=1 train_gpt.py +``` + +Best export-only ablation on the resulting checkpoint: + +```bash +RUN_ID=ar_export_rotq_mlpupdown_perlayer_had_long4800_fullcal_slide64 \ +INIT_MODEL_PATH=/workspace/parameter-golf/checkpoints/ar_xsa_base_long4800_seed314_final_model.pt \ +DATA_PATH=/workspace/parameter-golf/data/datasets/fineweb10B_sp1024/ \ +TOKENIZER_PATH=/workspace/parameter-golf/data/tokenizers/fineweb_1024_bpe.model \ +VOCAB_SIZE=1024 \ +BIGRAM_VOCAB_SIZE=3072 \ +BIGRAM_DIM=112 \ +WARMDOWN_ITERS=4000 \ +TARGET_MB=15.9 \ +SEED=314 \ +MAX_WALLCLOCK_SECONDS=0 \ +EVAL_FORCE_S64=1 \ +ROTQ_ENABLED=1 \ +ROTQ_FORCE=1 \ +ROTQ_MODE=hadamard \ +ROTQ_TARGETS=mlp_down,mlp_up \ +ROTQ_BLOCK_SIZES=128,256,512 \ +ROTQ_GROUP_SHARE=0 \ +ROTQ_SCORE_TENSORS=6 \ +ROTQ_SCORE_ROWS=256 \ +torchrun --standalone --nproc_per_node=1 train_gpt.py +``` + +## Compliance Notes + +- No validation leakage during training +- GPTQ calibration data is autoregressively self-generated by the trained model +- No network calls are required during evaluation +- Artifact remains under the decimal `16,000,000` byte cap +- This is explicitly a **non-record** submission because compute exceeds the main-track budget + +## Included Files + +- `train_gpt.py`: exact script used for the best export result in this folder +- `train.log`: recovered copy of the auto-generated log for the long training run +- `rotq_ablation.log`: recovered copy of the export-only ablation log +- `submission.json`: metadata for this non-record submission + +## Limitations + +- Single-seed evidence only +- Not reproduced on `8xH100 SXM` within the `600s` training budget +- The pod stopped before raw log sync completed, so the attached logs are recovered copies of the exact auto-generated log content captured from the pod session +- The ROTQ gain is real but small; the main value of this PR is the new modular export idea and the evidence that it composes with the current strong AR-GPTQ/XSA stack diff --git a/records/track_non_record_16mb/2026-04-01_ar_gptq_xsa_rotq_hadamard/requirements.txt b/records/track_non_record_16mb/2026-04-01_ar_gptq_xsa_rotq_hadamard/requirements.txt new file mode 100644 index 0000000000..68fda0b78a --- /dev/null +++ b/records/track_non_record_16mb/2026-04-01_ar_gptq_xsa_rotq_hadamard/requirements.txt @@ -0,0 +1,3 @@ +sentencepiece +zstandard +flash_attn_3 diff --git a/records/track_non_record_16mb/2026-04-01_ar_gptq_xsa_rotq_hadamard/rotq_ablation.log b/records/track_non_record_16mb/2026-04-01_ar_gptq_xsa_rotq_hadamard/rotq_ablation.log new file mode 100644 index 0000000000..74bd9e7c06 --- /dev/null +++ b/records/track_non_record_16mb/2026-04-01_ar_gptq_xsa_rotq_hadamard/rotq_ablation.log @@ -0,0 +1,93 @@ +starting ar_export_rotq_mlpdown_had256_long4800_fullcal_slide64 +world_size:1 grad_accum_steps:8 +sdp_backends:cudnn=False flash=True mem_efficient=False math=False +attention_kernel:flashattn3 +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:0 warmup_steps:0 max_wallclock_seconds:0.000 +seed:314 +step:0/0 val_loss:1.9122 val_bpb:1.1325 train_time:0ms step_avg:0.02ms +peak memory allocated: 1002 MiB reserved: 1242 MiB +ema:applying EMA weights +DIAGNOSTIC post_ema val_loss:1.9122 val_bpb:1.1325 eval_time:16429ms +Serialized model: 106289590 bytes +Code size: 119252 bytes +gptq:building non-banked model for Hessian collection... +gptq:generating autoregressive calibration data (64 seqs x 2048 tokens, temp=0.8)... +gptq:generated 64 sequences in 199.6s +gptq:collecting hessians from autoregressive data... +gptq:collected hessians for 68 layers (AR self-gen) +rotq:enabled groups:1 active:11 meta_bytes:264 +rotq:mlp_down:1536 group:mlp_down mode:hadamard block:256 score:1.564798e-01 +selective_prune: 4217740 ±1 candidates, unpruned=15.09MB target=15.9MB +selective_prune: already fits, no pruning needed +Serialized model int6+lzma: 15705748 bytes +Rotation metadata bytes (logical): 264 +Total submission size int6+lzma: 15825000 bytes +final_int6_roundtrip val_loss:1.9190 val_bpb:1.1365 eval_time:16283ms +final_int6_roundtrip_exact val_loss:1.91901076 val_bpb:1.13654627 +final_int6_sliding_window val_loss:1.8791 val_bpb:1.1129 stride:64 eval_time:595377ms +final_int6_sliding_window_exact val_loss:1.87910898 val_bpb:1.11291713 +final_int8_zlib_roundtrip_exact val_loss:1.87910898 val_bpb:1.11291713 +starting ar_export_rotq_mlpdown_perlayer_had_long4800_fullcal_slide64 +DIAGNOSTIC post_ema val_loss:1.9122 val_bpb:1.1325 eval_time:16381ms +Serialized model: 106289590 bytes +Code size: 119252 bytes +gptq:building non-banked model for Hessian collection... +gptq:generating autoregressive calibration data (64 seqs x 2048 tokens, temp=0.8)... +gptq:generated 64 sequences in 198.9s +gptq:collecting hessians from autoregressive data... +gptq:collected hessians for 68 layers (AR self-gen) +rotq:enabled groups:11 active:11 meta_bytes:264 +rotq:blocks.0.mlp.proj.weight group:mlp_down mode:hadamard block:128 score:1.569647e-01 +rotq:blocks.1.mlp.proj.weight group:mlp_down mode:hadamard block:512 score:1.561793e-01 +rotq:blocks.2.mlp.proj.weight group:mlp_down mode:hadamard block:128 score:1.568002e-01 +rotq:blocks.3.mlp.proj.weight group:mlp_down mode:hadamard block:128 score:1.558599e-01 +rotq:blocks.4.mlp.proj.weight group:mlp_down mode:hadamard block:512 score:1.558491e-01 +rotq:blocks.5.mlp.proj.weight group:mlp_down mode:hadamard block:128 score:1.564744e-01 +rotq:blocks.6.mlp.proj.weight group:mlp_down mode:hadamard block:512 score:1.560896e-01 +rotq:blocks.7.mlp.proj.weight group:mlp_down mode:hadamard block:128 score:1.559215e-01 +rotq:blocks.8.mlp.proj.weight group:mlp_down mode:hadamard block:256 score:1.554852e-01 +rotq:blocks.9.mlp.proj.weight group:mlp_down mode:hadamard block:512 score:1.563086e-01 +rotq:blocks.10.mlp.proj.weight group:mlp_down mode:hadamard block:128 score:1.571935e-01 +selective_prune: 4216026 ±1 candidates, unpruned=15.09MB target=15.9MB +selective_prune: already fits, no pruning needed +Serialized model int6+lzma: 15707228 bytes +Rotation metadata bytes (logical): 264 +Total submission size int6+lzma: 15826480 bytes +final_int6_roundtrip val_loss:1.9190 val_bpb:1.1365 eval_time:16266ms +final_int6_roundtrip_exact val_loss:1.91899953 val_bpb:1.13653961 +final_int6_sliding_window val_loss:1.8791 val_bpb:1.1129 stride:64 eval_time:592103ms +final_int6_sliding_window_exact val_loss:1.87909590 val_bpb:1.11290938 +final_int8_zlib_roundtrip_exact val_loss:1.87909590 val_bpb:1.11290938 +starting ar_export_rotq_mlpupdown_perlayer_had_long4800_fullcal_slide64 +rotq:blocks.2.mlp.fc.weight group:mlp_up mode:hadamard block:128 score:1.543067e-01 +rotq:blocks.3.mlp.fc.weight group:mlp_up mode:hadamard block:512 score:1.542510e-01 +rotq:blocks.4.mlp.fc.weight group:mlp_up mode:hadamard block:512 score:1.528345e-01 +rotq:blocks.5.mlp.fc.weight group:mlp_up mode:hadamard block:128 score:1.565601e-01 +rotq:blocks.6.mlp.fc.weight group:mlp_up mode:hadamard block:256 score:1.545500e-01 +rotq:blocks.7.mlp.fc.weight group:mlp_up mode:hadamard block:256 score:1.552076e-01 +rotq:blocks.8.mlp.fc.weight group:mlp_up mode:hadamard block:512 score:1.547106e-01 +rotq:blocks.9.mlp.fc.weight group:mlp_up mode:hadamard block:128 score:1.537057e-01 +rotq:blocks.10.mlp.fc.weight group:mlp_up mode:hadamard block:128 score:1.545073e-01 +rotq:blocks.0.mlp.proj.weight group:mlp_down mode:hadamard block:128 score:1.569647e-01 +rotq:blocks.1.mlp.proj.weight group:mlp_down mode:hadamard block:512 score:1.561793e-01 +rotq:blocks.2.mlp.proj.weight group:mlp_down mode:hadamard block:128 score:1.568002e-01 +rotq:blocks.3.mlp.proj.weight group:mlp_down mode:hadamard block:128 score:1.558599e-01 +rotq:blocks.4.mlp.proj.weight group:mlp_down mode:hadamard block:512 score:1.558491e-01 +rotq:blocks.5.mlp.proj.weight group:mlp_down mode:hadamard block:128 score:1.564744e-01 +rotq:blocks.6.mlp.proj.weight group:mlp_down mode:hadamard block:512 score:1.560896e-01 +rotq:blocks.7.mlp.proj.weight group:mlp_down mode:hadamard block:128 score:1.559215e-01 +rotq:blocks.8.mlp.proj.weight group:mlp_down mode:hadamard block:256 score:1.554852e-01 +rotq:blocks.9.mlp.proj.weight group:mlp_down mode:hadamard block:512 score:1.563086e-01 +rotq:blocks.10.mlp.proj.weight group:mlp_down mode:hadamard block:128 score:1.571935e-01 +selective_prune: 4208001 ±1 candidates, unpruned=15.09MB target=15.9MB +selective_prune: already fits, no pruning needed +Serialized model int6+lzma: 15706896 bytes +Rotation metadata bytes (logical): 506 +Total submission size int6+lzma: 15826148 bytes +final_int6_roundtrip val_loss:1.9190 val_bpb:1.1365 eval_time:16322ms +final_int6_roundtrip_exact val_loss:1.91899332 val_bpb:1.13653594 +final_int6_sliding_window val_loss:1.8791 val_bpb:1.1129 stride:64 eval_time:596342ms +final_int6_sliding_window_exact val_loss:1.87908996 val_bpb:1.11290586 +final_int8_zlib_roundtrip_exact val_loss:1.87908996 val_bpb:1.11290586 diff --git a/records/track_non_record_16mb/2026-04-01_ar_gptq_xsa_rotq_hadamard/submission.json b/records/track_non_record_16mb/2026-04-01_ar_gptq_xsa_rotq_hadamard/submission.json new file mode 100644 index 0000000000..c7f9207abb --- /dev/null +++ b/records/track_non_record_16mb/2026-04-01_ar_gptq_xsa_rotq_hadamard/submission.json @@ -0,0 +1,12 @@ +{ + "track": "non_record_16mb", + "date": "2026-04-01", + "name": "AR Self-Gen GPTQ + XSA-all + Per-Layer Hadamard ROTQ", + "author": "Arash", + "github_id": "vermissa0ss", + "val_bpb": 1.11290586, + "val_loss": 1.87908996, + "bytes_total": 15826148, + "hardware": "1xH100 80GB", + "blurb": "Non-record submission. Starts from the public AR self-generated GPTQ + XSA-all + BigramHash stack and adds rotation-aware GPTQ with per-layer Hadamard right-rotations on MLP matrices. Best export on a 1xH100/4800s run reached 1.11290586 sliding bpb at 15,826,148 bytes." +} diff --git a/records/track_non_record_16mb/2026-04-01_ar_gptq_xsa_rotq_hadamard/train.log b/records/track_non_record_16mb/2026-04-01_ar_gptq_xsa_rotq_hadamard/train.log new file mode 100644 index 0000000000..f49b43060d --- /dev/null +++ b/records/track_non_record_16mb/2026-04-01_ar_gptq_xsa_rotq_hadamard/train.log @@ -0,0 +1,80 @@ +model_params:27067484 +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:1 grad_accum_steps:8 +sdp_backends:cudnn=False flash=True mem_efficient=False math=False +attention_kernel:flashattn3 +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:4800.000 +seed:314 +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:6.9271 val_bpb:4.1026 train_time:0ms step_avg:0.02ms +step:1/20000 train_loss:6.9279 train_time:652ms step_avg:652.22ms +step:2/20000 train_loss:8.6225 train_time:1243ms step_avg:621.68ms +step:3/20000 train_loss:7.6539 train_time:1904ms step_avg:634.55ms +step:4/20000 train_loss:7.3319 train_time:2548ms step_avg:636.98ms +step:5/20000 train_loss:7.0682 train_time:3200ms step_avg:639.99ms +step:6/20000 train_loss:7.0074 train_time:3847ms step_avg:641.21ms +step:7/20000 train_loss:6.9721 train_time:4493ms step_avg:641.83ms +step:8/20000 train_loss:6.7644 train_time:5146ms step_avg:643.20ms +step:9/20000 train_loss:6.4777 train_time:5793ms step_avg:643.68ms +step:10/20000 train_loss:6.1204 train_time:6441ms step_avg:644.09ms +step:500/20000 train_loss:2.3192 train_time:325737ms step_avg:651.47ms +step:1000/20000 train_loss:2.2598 train_time:651864ms step_avg:651.86ms +step:1500/20000 train_loss:2.1287 train_time:978338ms step_avg:652.23ms +step:2000/20000 train_loss:2.0470 train_time:1305273ms step_avg:652.64ms +step:2500/20000 train_loss:2.0920 train_time:1632135ms step_avg:652.85ms +step:3000/20000 train_loss:2.0749 train_time:1959198ms step_avg:653.07ms +step:3500/20000 train_loss:2.0674 train_time:2286291ms step_avg:653.23ms +step:4000/20000 train_loss:2.1350 train_time:2613343ms step_avg:653.34ms +step:4000/20000 val_loss:2.0439 val_bpb:1.2105 train_time:2613392ms step_avg:653.35ms +step:4500/20000 train_loss:2.1269 train_time:2940377ms step_avg:653.42ms +step:5000/20000 train_loss:2.0362 train_time:3267311ms step_avg:653.46ms +step:5500/20000 train_loss:2.0406 train_time:3594311ms step_avg:653.51ms +step:6000/20000 train_loss:1.9402 train_time:3921190ms step_avg:653.53ms +step:6500/20000 train_loss:2.0445 train_time:4247900ms step_avg:653.52ms +swa:start step:6550 +late_qat:enabled step:6744 scale:0.1498 +step:7000/20000 train_loss:1.8408 train_time:4576477ms step_avg:653.78ms +step:7341/20000 val_loss:1.9137 val_bpb:1.1334 train_time:4800390ms step_avg:653.91ms +stopping_early: wallclock_cap train_time:4800390ms step:7341/20000 +peak memory allocated: 23002 MiB reserved: 23184 MiB +ema:applying EMA weights +DIAGNOSTIC post_ema val_loss:1.9121 val_bpb:1.1325 eval_time:16505ms +Serialized model: 106289590 bytes +Code size: 118057 bytes +gptq:building non-banked model for Hessian collection... +gptq:generating autoregressive calibration data (64 seqs x 2048 tokens, temp=0.8)... +gptq:generated 64 sequences in 204.7s +gptq:collecting hessians from autoregressive data... +gptq:collected hessians for 68 layers (AR self-gen) +selective_prune: 4217763 ±1 candidates, unpruned=15.13MB target=15.9MB +selective_prune: already fits, no pruning needed +Serialized model int6+lzma: 15751544 bytes +Rotation metadata bytes (logical): 0 +Total submission size int6+lzma: 15869601 bytes +final_int6_roundtrip val_loss:1.9191 val_bpb:1.1366 eval_time:20641ms +final_int6_roundtrip_exact val_loss:1.91911329 val_bpb:1.13660699 +final_int6_sliding_window val_loss:1.8792 val_bpb:1.1130 stride:64 eval_time:591478ms +final_int6_sliding_window_exact val_loss:1.87918563 val_bpb:1.11296252 +final_int8_zlib_roundtrip_exact val_loss:1.87918563 val_bpb:1.11296252 diff --git a/records/track_non_record_16mb/2026-04-01_ar_gptq_xsa_rotq_hadamard/train_gpt.py b/records/track_non_record_16mb/2026-04-01_ar_gptq_xsa_rotq_hadamard/train_gpt.py new file mode 100644 index 0000000000..251bc9c78c --- /dev/null +++ b/records/track_non_record_16mb/2026-04-01_ar_gptq_xsa_rotq_hadamard/train_gpt.py @@ -0,0 +1,2530 @@ +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_FLASH_ATTN3 = True +except ImportError: + flash_attn_3_func = None + _HAS_FLASH_ATTN3 = False +class Hyperparameters: + data_path = os.environ.get("DATA_PATH", "./data/datasets/fineweb10B_sp1024") + 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") + init_model_path = os.environ.get("INIT_MODEL_PATH", "") + 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)) + 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", 1.5)) + vocab_size = int(os.environ.get("VOCAB_SIZE", 1024)) + 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", 3.0)) + 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)) + eval_force_s64 = bool(int(os.environ.get("EVAL_FORCE_S64", "1"))) + 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", 0.04)) + adam_wd = float(os.environ.get("ADAM_WD", 0.04)) + qat_enabled = bool(int(os.environ.get("QAT_ENABLED", "0"))) + bigram_vocab_size = int(os.environ.get("BIGRAM_VOCAB_SIZE", 2048)) + 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", "1"))) + 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) + # GPTQ calibration + gptq_calib_batches = int(os.environ.get("GPTQ_CALIB_BATCHES", 256)) + gptq_block_size = int(os.environ.get("GPTQ_BLOCK_SIZE", 128)) + gptq_calib_num_seqs = int(os.environ.get("GPTQ_CALIB_NUM_SEQS", 64)) + gptq_calib_seq_len = int(os.environ.get("GPTQ_CALIB_SEQ_LEN", 2048)) + gptq_calib_batch_size = int(os.environ.get("GPTQ_CALIB_BATCH_SIZE", 8)) + gptq_calib_temperature = float(os.environ.get("GPTQ_CALIB_TEMPERATURE", 0.8)) + # Rotation-aware GPTQ + rotq_enabled = bool(int(os.environ.get("ROTQ_ENABLED", "0"))) + rotq_mode = os.environ.get("ROTQ_MODE", "auto") + rotq_block_sizes = os.environ.get("ROTQ_BLOCK_SIZES", "128,256,512") + rotq_targets = os.environ.get("ROTQ_TARGETS", "attn,mlp") + rotq_global_seed = int(os.environ.get("ROTQ_GLOBAL_SEED", 12345)) + rotq_num_seeds = int(os.environ.get("ROTQ_NUM_SEEDS", 8)) + rotq_num_reflections = int(os.environ.get("ROTQ_NUM_REFLECTIONS", 2)) + rotq_search_steps = int(os.environ.get("ROTQ_SEARCH_STEPS", 16)) + rotq_group_share = bool(int(os.environ.get("ROTQ_GROUP_SHARE", "1"))) + rotq_force = bool(int(os.environ.get("ROTQ_FORCE", "0"))) + rotq_log_details = bool(int(os.environ.get("ROTQ_LOG_DETAILS", "1"))) + rotq_score_tensors = int(os.environ.get("ROTQ_SCORE_TENSORS", 6)) + rotq_score_rows = int(os.environ.get("ROTQ_SCORE_ROWS", 256)) + +# --- 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,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) + + +def attn_func_bthd(q: Tensor, k: Tensor, v: Tensor) -> Tensor: + if _HAS_FLASH_ATTN3: + return flash_attn_3_func(q, k, v, causal=True) + q_s = q.transpose(1, 2) + k_s = k.transpose(1, 2) + v_s = v.transpose(1, 2) + y = F.scaled_dot_product_attention(q_s, k_s, v_s, is_causal=True, enable_gqa=True) + return y.transpose(1, 2).contiguous() + +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] + y = attn_func_bthd(q, k, v) + 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, + ): + 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)) + # 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: + 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, + 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_logits(self, input_ids: Tensor) -> Tensor: + """Return logits (bsz, seq_len, vocab) without computing loss.""" + 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: + 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, + 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) + if self.tie_embeddings: + logits_proj = F.linear(x, self.tok_emb.weight) + else: + logits_proj = self.lm_head(x) + return self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + +# --- 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 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') + def make_hook(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 + h = module.register_forward_hook(make_hook(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 quantize_int6_per_row(t: Tensor, clip_range: int = 31) -> tuple[Tensor, Tensor]: + t32 = t.float() + if t32.ndim == 2: + best_q, best_s, best_err = None, None, float('inf') + for pct in [0.9990, 0.9995, 0.9999, 0.99999, 1.0]: + if pct < 1.0: + row_clip = torch.quantile(t32.abs(), pct, dim=1) + else: + row_clip = t32.abs().amax(dim=1) + s = (row_clip / clip_range).clamp_min(1.0 / clip_range).to(torch.float16) + q = torch.clamp(torch.round(t32 / s.float()[:, None]), -clip_range, clip_range).to(torch.int8) + recon = q.float() * s.float()[:, None] + err = (t32 - recon).pow(2).mean().item() + if err < best_err: + best_q, best_s, best_err = q, s, err + return best_q, best_s + amax = t32.abs().max().item() + scale = torch.tensor(amax / clip_range if amax > 0 else 1.0, dtype=torch.float16) + q = torch.clamp(torch.round(t32 / scale.float()), -clip_range, clip_range).to(torch.int8) + return q, scale + +def quantize_int6_gptq(weight, hessian=None, clip_range=31, block_size=128): + """Full GPTQ: Hessian-aware int6 quantization with Cholesky error compensation. + If hessian is None, falls back to percentile search.""" + t32 = weight.float() + if t32.ndim != 2 or hessian is None: + return _quantize_int6_percentile(t32, clip_range) + rows, cols = t32.shape + H = hessian.float().clone() + H = 0.5 * (H + H.T) + dead = torch.diag(H) == 0 + H[dead, dead] = 1 + base_damp = float((0.01 * torch.mean(torch.diag(H)).clamp_min(1e-6)).item()) + 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 = None + eye = torch.eye(cols, dtype=H.dtype, device=H.device) + for mul in (1.0, 10.0, 100.0, 1000.0, 10000.0): + try: + H_work = H + eye * (base_damp * mul) + H_work = 0.5 * (H_work + H_work.T) + chol = torch.linalg.cholesky(H_work) + Hinv = torch.cholesky_inverse(chol) + Hinv = torch.linalg.cholesky(Hinv, upper=True) + break + except torch._C._LinAlgError: + continue + if Hinv is None: + return _quantize_int6_percentile(t32, clip_range) + best_q = None; best_scale = None; best_err = float('inf') + for pct in [0.9990, 0.9995, 0.9999, 0.99999, 1.0]: + if pct < 1.0: + row_clip = torch.quantile(t32.abs(), pct, dim=1) + else: + row_clip = t32.abs().amax(dim=1) + s = (row_clip / clip_range).clamp_min(1.0 / clip_range).to(torch.float16) + 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:] + recon = Q.float() * sf[:, None] + mse = (W - recon).pow(2).mean().item() + if mse < best_err: + best_q, best_scale, best_err = Q, s, mse + best_q = best_q[:, inv_perm] + return best_q, best_scale + +def _quantize_int6_percentile(t32, clip_range=31): + """Fallback: percentile search (for 1D or no-Hessian cases).""" + if t32.ndim == 2: + best_q, best_s, best_err = None, None, float('inf') + for pct in [0.9990, 0.9995, 0.9999, 0.99999, 1.0]: + if pct < 1.0: + row_clip = torch.quantile(t32.abs(), pct, dim=1) + else: + row_clip = t32.abs().amax(dim=1) + s = (row_clip / clip_range).clamp_min(1.0 / clip_range).to(torch.float16) + q = torch.clamp(torch.round(t32 / s.float()[:, None]), -clip_range, clip_range).to(torch.int8) + recon = q.float() * s.float()[:, None] + err = (t32 - recon).pow(2).mean().item() + if err < best_err: + best_q, best_s, best_err = q, s, err + return best_q, best_s + amax = t32.abs().max().item() + scale = torch.tensor(amax / clip_range if amax > 0 else 1.0, dtype=torch.float16) + q = torch.clamp(torch.round(t32 / scale.float()), -clip_range, clip_range).to(torch.int8) + return q, scale + +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] + y = attn_func_bthd(q, k, v) + 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 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.""" + 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') + def make_hook(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 + h = module.register_forward_hook(make_hook(param_name)) + hooks.append(h) + 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, + args: Hyperparameters | None = None, + log0=None, +): + 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] = {} + rot_meta: dict[str, dict[str, object]] = {} + rot_scores: list[dict[str, object]] = [] + if args is not None and args.rotq_enabled: + rotatables = iter_rotatable_tensors(state_dict, args) + groups: dict[str, list[dict[str, object]]] = {} + for item in rotatables: + key = item["share_key"] if args.rotq_group_share else str(item["name"]) + groups.setdefault(key, []).append(item) + for key, items in groups.items(): + tensors = [item["tensor"] for item in items] + group_name = str(items[0]["group_name"]) + chosen_meta, chosen_score = choose_rotation_for_group(group_name, tensors, args) + if chosen_meta.get("mode") != "identity": + for item in items: + rot_meta[str(item["name"])] = chosen_meta + rot_scores.append( + { + "key": key, + "group_name": group_name, + "mode": chosen_meta.get("mode", "identity"), + "block_size": int(chosen_meta.get("block_size", 1)), + "score": float(chosen_score), + } + ) + if log0 is not None and args.rotq_log_details: + rot_bytes = estimate_rot_meta_bytes(rot_meta) + log0(f"rotq:enabled groups:{len(groups)} active:{len(rot_meta)} meta_bytes:{rot_bytes}") + for row in rot_scores: + log0( + f"rotq:{row['key']} group:{row['group_name']} mode:{row['mode']} " + f"block:{row['block_size']} score:{row['score']:.6e}", + console=False, + ) + 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 in int6_cats and t.ndim >= 1: + cr = 31 # int6 for all weights + H = hessians.get(name) if hessians else None + rot = rot_meta.get(name) + if rot is not None and t.ndim == 2: + t = apply_right_rotation(t.float(), rot).contiguous() + if H is not None: + H = transform_hessian_right(H.float(), rot).contiguous() + if H is not None: + q, s = quantize_int6_gptq(t, hessian=H, clip_range=cr) + else: + q, s = quantize_int6_per_row(t, clip_range=cr) + result[name + ".q"] = q + result[name + ".scale"] = s + meta[name] = {"type": "int6"} + else: + q, s = quantize_float_tensor(t) + result[name + ".q"] = q + result[name + ".scale"] = s + meta[name] = {"type": "int8"} + return result, meta, rot_meta, rot_scores +def dequantize_mixed_int6( + result: dict[str, Tensor], + meta: dict[str, object], + template_sd: dict[str, Tensor], + rot_meta: dict[str, dict[str, object]] | None = None, +) -> 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) + rot = rot_meta.get(name) if rot_meta is not None else None + if rot is not None and out[name].ndim == 2: + out[name] = invert_right_rotation(out[name].float(), rot).to(orig_dtype) + return out + +# --- Rotation-aware GPTQ helpers --- + +def _parse_csv_ints(text: str) -> list[int]: + return [int(part) for part in text.split(",") if part.strip()] + + +def _parse_csv_set(text: str) -> set[str]: + return {part.strip() for part in text.split(",") if part.strip()} + + +def _stable_hash64(text: str) -> int: + h = 1469598103934665603 + for b in text.encode("utf-8"): + h ^= b + h = (h * 1099511628211) & 0xFFFFFFFFFFFFFFFF + return h + + +def _classify_rotq_tensor(name: str) -> tuple[str, str] | None: + if name == "tok_emb.weight": + return "embed", "embed" + if name.endswith("lm_head.weight"): + return "head", "head" + if name.endswith(".attn.c_q.weight") or name.endswith(".attn.c_k.weight") or name.endswith(".attn.c_v.weight"): + return "attn", "attn_in" + if name.endswith(".attn.proj.weight"): + return "attn", "attn_out" + if name.endswith(".mlp.fc.weight"): + return "mlp", "mlp_up" + if name.endswith(".mlp.proj.weight"): + return "mlp", "mlp_down" + return None + + +def iter_rotatable_tensors(state_dict: dict[str, Tensor], args: Hyperparameters) -> list[dict[str, object]]: + targets = _parse_csv_set(args.rotq_targets) + out: list[dict[str, object]] = [] + for name, tensor in state_dict.items(): + cls = _classify_rotq_tensor(name) + if cls is None: + continue + target, group_name = cls + if (target not in targets and group_name not in targets) or tensor.ndim != 2 or not tensor.is_floating_point(): + continue + if tensor.numel() <= 65536: + continue + out.append( + { + "name": name, + "tensor": tensor, + "group_name": group_name, + "share_key": f"{group_name}:{tensor.shape[1]}", + } + ) + return out + + +def fwht_lastdim_blockwise(x: Tensor, block_size: int) -> Tensor: + if block_size <= 1: + return x + if x.shape[-1] % block_size != 0: + raise ValueError(f"Last dim {x.shape[-1]} must be divisible by block_size={block_size}") + y = x.reshape(-1, x.shape[-1] // block_size, block_size).clone() + h = 1 + while h < block_size: + y = y.view(y.shape[0], y.shape[1], block_size // (2 * h), 2, h) + a = y[..., 0, :].clone() + b = y[..., 1, :].clone() + y[..., 0, :] = a + b + y[..., 1, :] = a - b + y = y.view(y.shape[0], y.shape[1], block_size) + h *= 2 + return (y / math.sqrt(block_size)).reshape_as(x) + + +def apply_householder_stack_lastdim_blockwise(x: Tensor, vecs: Tensor) -> Tensor: + if vecs.ndim != 2: + raise ValueError(f"vecs must be 2D, got shape {tuple(vecs.shape)}") + block_size = vecs.shape[-1] + if x.shape[-1] % block_size != 0: + raise ValueError(f"Last dim {x.shape[-1]} must be divisible by block_size={block_size}") + y = x.reshape(-1, x.shape[-1] // block_size, block_size).clone() + vecs_n = F.normalize(vecs.float(), dim=-1).to(dtype=y.dtype, device=y.device) + for v in vecs_n: + proj = (y * v).sum(dim=-1, keepdim=True) + y = y - 2.0 * proj * v + return y.reshape_as(x) + + +def _valid_rotq_block_sizes(width: int, args: Hyperparameters) -> list[int]: + out = [] + for block_size in _parse_csv_ints(args.rotq_block_sizes): + if block_size > 1 and width % block_size == 0 and (block_size & (block_size - 1)) == 0: + out.append(block_size) + return out + + +def _make_sign_mask(num_blocks: int, block_size: int, meta: dict[str, object], device: torch.device, dtype: torch.dtype) -> Tensor: + seed = _stable_hash64( + f"{meta.get('group_name','')}/{meta.get('mode','')}/{meta.get('seed',0)}/{meta.get('block_size',block_size)}" + ) + gen = torch.Generator(device="cpu") + gen.manual_seed(seed) + bits = torch.randint(0, 2, (num_blocks, block_size), generator=gen, dtype=torch.int8) + return bits.to(dtype=torch.float32).mul_(2).sub_(1).to(device=device, dtype=dtype) + + +def apply_right_rotation(W: Tensor, rot_meta: dict[str, object]) -> Tensor: + mode = str(rot_meta.get("mode", "identity")) + if mode == "identity": + return W + if mode == "householder_shared": + base_mode = str(rot_meta.get("base_mode", "identity")) + base_meta = dict(rot_meta) + base_meta["mode"] = base_mode + Wr = apply_right_rotation(W, base_meta) if base_mode != "identity" else W + return apply_householder_stack_lastdim_blockwise(Wr, rot_meta["vecs"]) + block_size = int(rot_meta["block_size"]) + Wr = fwht_lastdim_blockwise(W, block_size) + if mode == "signed_hadamard": + y = Wr.reshape(-1, Wr.shape[-1] // block_size, block_size) + y = y * _make_sign_mask(y.shape[1], block_size, rot_meta, y.device, y.dtype).unsqueeze(0) + return y.reshape_as(Wr) + if mode == "hadamard": + return Wr + raise ValueError(f"Unknown ROTQ mode: {mode}") + + +def invert_right_rotation(Wr: Tensor, rot_meta: dict[str, object]) -> Tensor: + mode = str(rot_meta.get("mode", "identity")) + if mode == "identity": + return Wr + if mode == "householder_shared": + vecs = rot_meta["vecs"] + W = apply_householder_stack_lastdim_blockwise(Wr, torch.flip(vecs, dims=(0,))) + base_mode = str(rot_meta.get("base_mode", "identity")) + if base_mode == "identity": + return W + base_meta = dict(rot_meta) + base_meta["mode"] = base_mode + return invert_right_rotation(W, base_meta) + return apply_right_rotation(Wr, rot_meta) + + +def transform_hessian_right(H: Tensor, rot_meta: dict[str, object]) -> Tensor: + tmp = apply_right_rotation(H, rot_meta) + return apply_right_rotation(tmp.T.contiguous(), rot_meta).T.contiguous() + + +def fake_quantize_tensor_int6(W: Tensor, clip_range: int = 31) -> Tensor: + q, s = quantize_int6_per_row(W, clip_range=clip_range) + return (q.float() * s.float().view(q.shape[0], *([1] * (q.ndim - 1)))).to(dtype=W.dtype) + + +def estimate_rot_meta_bytes(rot_meta: dict[str, dict[str, object]]) -> int: + total = 0 + for meta in rot_meta.values(): + total += len(str(meta.get("mode", ""))) + len(str(meta.get("group_name", ""))) + for key in ("block_size", "seed", "num_reflections"): + if key in meta: + total += 4 + vecs = meta.get("vecs") + if isinstance(vecs, Tensor): + total += int(vecs.numel()) * int(vecs.element_size()) + return total + + +def _sample_rotq_group_tensors(tensors: list[Tensor], max_tensors: int, max_rows: int) -> list[Tensor]: + if not tensors: + return [] + if len(tensors) <= max_tensors: + chosen = tensors + else: + # Spread samples across the group so shared rotations don't overfit early layers. + idxs = torch.linspace(0, len(tensors) - 1, steps=max_tensors).round().to(dtype=torch.int64).tolist() + chosen = [tensors[i] for i in idxs] + out: list[Tensor] = [] + for tensor in chosen: + if tensor.ndim != 2 or tensor.shape[0] <= max_rows: + out.append(tensor) + else: + stride = max(tensor.shape[0] // max_rows, 1) + out.append(tensor[::stride][:max_rows].contiguous()) + return out + + +def _score_rotation_candidate_one(W: Tensor, rot_meta: dict[str, object]) -> float: + Wf = W.float() + Wr = apply_right_rotation(Wf, rot_meta) + W_fake = fake_quantize_tensor_int6(Wr) + denom = float(Wf.pow(2).mean().item()) + 1e-8 + qerr = float((W_fake - Wr).pow(2).mean().item()) / denom + l4 = float(Wr.abs().pow(4).mean().item()) / (denom * denom) + clip = float(torch.quantile(Wr.abs().reshape(-1), 0.999).item()) if Wr.numel() else 0.0 + tail = float(torch.relu(Wr.abs() - clip).pow(2).mean().item()) / denom if clip > 0 else 0.0 + return qerr + 0.05 * l4 + 0.01 * tail + + +def score_rotation_candidate(tensors: list[Tensor], rot_meta: dict[str, object], args: Hyperparameters) -> float: + sample = _sample_rotq_group_tensors(tensors, max(args.rotq_score_tensors, 1), max(args.rotq_score_rows, 1)) + return sum(_score_rotation_candidate_one(t, rot_meta) for t in sample) / max(len(sample), 1) + + +def choose_rotation_for_group(group_name: str, tensors: list[Tensor], args: Hyperparameters) -> tuple[dict[str, object], float]: + width = int(tensors[0].shape[1]) + block_sizes = _valid_rotq_block_sizes(width, args) + best_meta: dict[str, object] = {"mode": "identity", "group_name": group_name, "block_size": 1} + best_score = score_rotation_candidate(tensors, best_meta, args) + best_explicit_meta: dict[str, object] | None = None + best_explicit_score = float("inf") + modes = [args.rotq_mode] if args.rotq_mode != "auto" else ["identity", "hadamard", "signed_hadamard", "householder_shared"] + for mode in modes: + if mode == "identity": + continue + if mode in {"hadamard", "signed_hadamard"}: + for block_size in block_sizes: + seeds = [args.rotq_global_seed] if mode == "hadamard" else [args.rotq_global_seed + i for i in range(args.rotq_num_seeds)] + for seed in seeds: + meta = {"mode": mode, "group_name": group_name, "block_size": block_size, "seed": seed} + score = score_rotation_candidate(tensors, meta, args) + if score < best_explicit_score: + best_explicit_meta, best_explicit_score = meta, score + if score < best_score: + best_meta, best_score = meta, score + elif mode == "householder_shared" and block_sizes: + init_meta = dict(best_meta) if best_meta.get("mode") != "identity" else { + "mode": "hadamard", + "group_name": group_name, + "block_size": block_sizes[0], + "seed": args.rotq_global_seed, + } + block_size = int(init_meta["block_size"]) + gen = torch.Generator(device="cpu") + gen.manual_seed(_stable_hash64(f"{group_name}/{args.rotq_global_seed}/{block_size}")) + vecs = torch.randn(args.rotq_num_reflections, block_size, generator=gen, dtype=torch.float32) / math.sqrt(block_size) + vecs = nn.Parameter(vecs) + opt = torch.optim.Adam([vecs], lr=0.05) + base_mode = str(init_meta.get("mode", "hadamard")) + for _ in range(max(args.rotq_search_steps, 1)): + opt.zero_grad(set_to_none=True) + meta = { + "mode": "householder_shared", + "group_name": group_name, + "block_size": block_size, + "num_reflections": args.rotq_num_reflections, + "base_mode": base_mode, + "seed": int(init_meta.get("seed", args.rotq_global_seed)), + "vecs": vecs, + } + loss = 0.0 + for tensor in _sample_rotq_group_tensors(tensors, max(args.rotq_score_tensors, 1), max(args.rotq_score_rows, 1)): + Wr = apply_right_rotation(tensor.float(), meta) + W_fake = fake_quantize_tensor_int6(Wr) + loss = loss + (W_fake - Wr).pow(2).mean() + loss.backward() + opt.step() + meta = { + "mode": "householder_shared", + "group_name": group_name, + "block_size": block_size, + "num_reflections": args.rotq_num_reflections, + "base_mode": base_mode, + "seed": int(init_meta.get("seed", args.rotq_global_seed)), + "vecs": vecs.detach().cpu().to(dtype=torch.float16).contiguous(), + } + score = score_rotation_candidate(tensors, meta, args) + if score < best_explicit_score: + best_explicit_meta, best_explicit_score = meta, score + if score < best_score: + best_meta, best_score = meta, score + if args.rotq_force and args.rotq_mode != "auto" and best_explicit_meta is not None: + return best_explicit_meta, best_explicit_score + return best_meta, best_score + +# --- 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, + ).to(device).bfloat16() + if args.init_model_path: + init_state = torch.load(args.init_model_path, map_location="cpu") + base_model.load_state_dict(init_state, strict=False) + args.warmup_steps = 0 + args.iterations = 0 + args.max_wallclock_seconds = 0.0 + log0(f"init_model_path:{args.init_model_path} mode:export_only") + # 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) + 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.adam_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_kernel:{'flashattn3' if _HAS_FLASH_ATTN3 else 'torch_sdpa_fallback'}") + 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 lr_mul(step: int, elapsed_ms: float) -> float: + 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" + ) + 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, + ) + # Autoregressive self-generated calibration (no external data) + log0( + "gptq:generating autoregressive calibration data " + f"({args.gptq_calib_num_seqs} seqs x {args.gptq_calib_seq_len} tokens, temp={args.gptq_calib_temperature})..." + ) + base_model.load_state_dict(export_sd, strict=False) + t_gen = time.perf_counter() + ar_tokens = generate_autoregressive_calib( + base_model, + device, + num_seqs=args.gptq_calib_num_seqs, + seq_len=args.gptq_calib_seq_len, + vocab_size=args.vocab_size, + temperature=args.gptq_calib_temperature, + batch_size=args.gptq_calib_batch_size, + seed=args.seed, + ) + log0(f"gptq:generated {len(ar_tokens)} sequences in {time.perf_counter()-t_gen:.1f}s") + log0("gptq:collecting hessians from autoregressive data...") + hessians = collect_hessians_from_tokens(hessian_model, ar_tokens, device) + log0(f"gptq:collected hessians for {len(hessians)} layers (AR self-gen)") + del ar_tokens + del hessian_model + torch.cuda.empty_cache() + quant_result, quant_meta, rot_meta, rot_scores = mixed_quantize_int6( + unbanked_sd, + {"mlp", "attn"}, + hessians=hessians, + args=args, + log0=log0, + ) + # 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, "r": rot_meta}, buf) + return len(lzma.compress(buf.getvalue(), preset=9)) + 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, "r": rot_meta}, quant_buf) + quant_raw = quant_buf.getvalue() + quant_blob = lzma.compress(quant_raw, preset=9) + 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")) + rot_bytes = estimate_rot_meta_bytes(rot_meta) + log0(f"Serialized model int6+lzma: {quant_file_bytes} bytes") + log0(f"Rotation metadata bytes (logical): {rot_bytes}") + log0(f"Total submission size int6+lzma: {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(lzma.decompress(quant_blob_disk)), + map_location="cpu", + ) + deq_unbanked = dequantize_mixed_int6(quant_state["w"], quant_state["m"], unbanked_sd, quant_state.get("r")) + # 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, + ).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_force_s64 and 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}") + if distributed: + dist.destroy_process_group() +if __name__ == "__main__": + main()