Non-record: Fixed Bank QAT + XSA5 + Label Smoothing (1.1352)#667
Open
suchitj2702 wants to merge 1 commit intoopenai:mainfrom
Open
Non-record: Fixed Bank QAT + XSA5 + Label Smoothing (1.1352)#667suchitj2702 wants to merge 1 commit intoopenai:mainfrom
suchitj2702 wants to merge 1 commit intoopenai:mainfrom
Conversation
Non-record experimental submission exploring: - STE int6 fake-quantization fix for bank parameters (QAT was dead code) - XSA expanded to last 5 layers - Label smoothing 0.05 - TTT LR/momentum tuning Result: 1.1352 BPB (worse than SOTA 1.1194). Key findings: - QAT recompilation too expensive (~50s + 5ms/step overhead) - Label smoothing counterproductive on compute-limited model - XSA5 neutral-to-negative vs XSA4 - Original TTT hyperparameters were better 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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 experimental submission exploring three changes on top of the LeakyReLU + Legal TTT + Parallel Muon stack:
GPT.forward()for all bank parameters. The SOTA's QAT is dead code because bank params bypassCastedLinearandtorch.compileconstant-folds the_qat_enabledflag.Result: 1.1352 BPB — does not beat SOTA (1.1194). Submitted as a non-record with findings.
Key Findings
Test plan
🤖 Generated with Claude Code