12L INT4 bQAT + EMA Fix + Deterministic QAT — val_bpb ~1.165

[SoHarshh]

12-layer INT4 MLP+bigram QAT with EMA fix and deterministic QAT timing.

val_bpb: ~1.165 | 15.90 MB | 8×H100 SXM | seed 1

Key contributions:

• INT4 bigram QAT (novel): first competition submission to quantize the bigram table below INT6, saves ~370KB enabling 12L in 16MB
• EMA reset at QAT activation: eliminates quantization degradation (+0.002 BPB vs +0.193 BPB without fix)
• Deterministic wallclock QAT trigger (LATE_QAT_FRAC=0.65): removes seed-to-seed QAT timing variance on multi-GPU runs

Check README.md for full details.

[MatoTeziTanka]

Community Review — 12L INT4 bQAT + EMA Fix + Deterministic QAT — val_bpb ~1.165

BPB: 0.002 (cache parse — may be delta/std, not val_bpb; check PR title) | Compliance: LOOKS CLEAN — score-first-per-chunk TTT (legal #1416/#1423 pattern)

What I found in the code (head SHA 8f803e96c525, file records/track_10min_16mb/2026-03-28_12L_INT4_bQAT_EMAfix/train_gpt.py):

The TTT path at line 1014 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 0.05s, dim=512, layers=10, vocab=1024, code=78582 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 0.05s, dim=512, layers=10, vocab=1024, code=78582 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.