Non-record: QAT Int5/Int6 on #1 architecture (1.14476 BPB)#306
Open
xuafeng wants to merge 1 commit intoopenai:mainfrom
Open
Non-record: QAT Int5/Int6 on #1 architecture (1.14476 BPB)#306xuafeng wants to merge 1 commit intoopenai:mainfrom
xuafeng wants to merge 1 commit intoopenai:mainfrom
Conversation
STE fake-quantization during training (int5 MLP, int6 attn) on top of thwu1's openai#1 entry. Best result: 1.14476 BPB (seed 1337). Key finding: post-training quantization + SWA outperforms QAT by ~0.002 BPB — quantization noise acts as beneficial regularization. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Non-record submission exploring QAT (Quantization-Aware Training) with STE fake-quantization on top of thwu1's #1 entry.
Best result: val_bpb = 1.14476 (seed 1337, 8xH100 SXM, 600s)
Key Finding
Post-training quantization + SWA outperforms QAT by ~0.002 BPB. The quantization noise from int5/int6 post-training quantization acts as beneficial regularization that QAT removes.
Files
records/track_non_record_16mb/2026-03-21_QAT_Int5_TTT_LoRA_8xH100/README.md— Detailed writeup with ablationssubmission.json— Metadatatrain_gpt.py— Reproducible scriptReproducibility
🤖 Generated with Claude Code