adaLN_recurrence [val_bpb=1.255 on 4 x H100]#1944
Open
dmitriymyan1 wants to merge 1 commit intoopenai:mainfrom
Open
adaLN_recurrence [val_bpb=1.255 on 4 x H100]#1944dmitriymyan1 wants to merge 1 commit intoopenai:mainfrom
dmitriymyan1 wants to merge 1 commit intoopenai:mainfrom
Conversation
Add adaLN (adaptive layer norm) conditioned on recurrence iteration to the Parallel Residuals + Mini Depth Recurrence baseline. Allows weight-tied recurrent layers to distinguish first vs second pass with ~zero compute overhead (~6.6K extra parameters). Early result: val_bpb 1.2551 on 4xH100/600s (half compute, only ~400 recurrent steps before wallclock cap). Co-Authored-By: Claude Opus 4.6 <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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
Early Result
Smoke-test on 4×H100 / 600s (50% of submission compute): val_bpb 1.2551 (val_loss 2.1193 nats), 15.26 MB quantized artifact. Only ~400 recurrent training steps ran before wallclock cap — loss curve still descending cleanly at cutoff. Full 8×H100 run pending.
Files
train_gpt.py— training script with adaLN support (FILM_ENABLED=1)README.md— approach description and reproducibility instructionsrequirements.txt— dependencies (addsbrotli)