Non-record: Mixed-Int6 LZMA9 B3072 Warm5000

[sabdulmajid]

Summary

Adds a non-record unlimited-compute 16MB submission: Mixed-Int6 LZMA9 B3072 Warm5000.

This is not a 10-minute record attempt, rather a 16.1h single-GPU run using the established EMA + XSA(last-4) + BigramHash3072 + LeakyReLU^2 flat-transformer stack, then exports the preserved raw checkpoint with broad mixed-int6 over mlp;attn;embed and LZMA9 extreme compression.

Result

• val_bpb: 1.20289664
• val_loss: 1.99963255
• pre-quant sliding BPB: 1.16618894
• artifact bytes: 15,991,188

This beats the listed 4-hour non-record baseline but does not beat the current 1-bit non-record result or the 10-minute SOTA -- was a fun little experiment I trained and ran on limited compute (as a student in college 😃)

Training: NVIDIA RTX A4500, 20GB VRAM
Export & eval: NVIDIA GeForce RTX 3050, 8GB VRAM

[MatoTeziTanka]

Community Review — Non-record: Mixed-Int6 LZMA9 B3072 Warm5000

BPB: 1.2029 | Compliance: LOOKS CLEAN — score-first-per-chunk TTT (legal #1416/#1423 pattern)

What I found in the code (head SHA 05b511b19db7, file records/track_non_record_16mb/2026-04-07_MixedInt6_LZMA9_B3072_Warm5000/train_gpt.py):

The TTT path at line 1329 implements the score-first-per-chunk pattern: each chunk is scored under torch.no_grad() / inference_mode() before the base_model.train() + SGD adaptation runs on that same chunk, with an is_last_chunk guard so the final chunk gets no adaptation pass. This is the structural shape the legal frontier uses (PRs #1416 erichroepke, #1423 aryanbhosale).

Per Issue #402 and Issue #677, TTT is legal when each token is scored before the adapter updates on it, and that's what the code does here — chunk ci is scored under weights adapted only on chunks 0..ci-1. No prequant_ttt_adapt_adamw(val_tokens, ...) multi-epoch fine-tune, no scored-region SLOT, no target-in-key n-gram cache.

CPU smoke test (CT2038 proteus-engine, 2026-04-11): import OK in 0.06s, dim=512, layers=11, vocab=1024, code=109840 B, SMOKE_TEST_PASS

Verdict: LOOKS CLEAN.

Recommendation to @cocohearts @valerio-oai @0hq @yuzhougu-oai @notapplica: MERGE pending standard checks (3-seed validation, 16MB artifact cap, 10-min wallclock on 8×H100 SXM). The compliance picture matches the legal reference frontier and no flags were raised by the classification pass.

Auto-classification caveat: this review was drafted by the AST-based classifier against a template derived from manually-reviewed cluster PRs (#1420, #1450, #1487, #1541, #1529, #1533, #1518). If I've misread a subtlety in your eval path — e.g., multi-epoch TTT that I mistook for single-pass, or a target-in-key lookup I missed in a helper function — please flag it and I'll re-run the audit manually.

---

Reviewed by @MatoTeziTanka — The Agora. CPU smoke test (CT2038 proteus-engine, 2026-04-11): import OK in 0.06s, dim=512, layers=11, vocab=1024, code=109840 B, SMOKE_TEST_PASS. Classification via deterministic AST-based classify_prs.py (pattern bank derived from ~65 manually-reviewed PRs earlier in the 2026-04-11 sweep). This review was auto-drafted from a template and spot-checked before posting — if the template misread your code, please call it out so I can iterate the classifier.