Record: PR549 + MiLe decay + 8-bit Muon + 1.04x LR + Cache+Backout — val_bpb 1.1176#703
Open
Gusanidas wants to merge 1 commit intoopenai:mainfrom
Open
Record: PR549 + MiLe decay + 8-bit Muon + 1.04x LR + Cache+Backout — val_bpb 1.1176#703Gusanidas wants to merge 1 commit intoopenai:mainfrom
Gusanidas wants to merge 1 commit intoopenai:mainfrom
Conversation
…val_bpb 1.1176 Four orthogonal improvements on PR openai#549 (LeakyReLU² + Legal TTT + Parallel Muon): - MiLe loss: entropy-weighted token loss with γ=1.1 decaying to 0 during warmdown - 8-bit Muon momentum: blockwise symmetric int8 quantization of momentum buffers - 1.04x LR boost: all learning rates scaled by 1.04x - Cache+Backout: cache layer 7 state, late attention reads cached context, subtract backout_lambda * cache from final output Tested on 4xB200 with wall clock 903s to match equivalent 8xH100 compute. 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
Four orthogonal improvements on PR #549 (LeakyReLU² + Legal TTT + Parallel Muon):
((1-exp(-entropy))^γ)with γ=1.1 decaying to 0 during warmdown. Focuses training on harder tokens early, then reverts to standard CE.x = x - λ·x_cachewhere λ is a learned scalar (init 0.1).Tested on 4xB200 with wall clock 903s to match equivalent 8xH100 compute.
Pending to be tested on 8xH100 SXM.
Test plan
🤖 Generated with Claude Code