60% more efficient autoresearch via better training analysis#353
Open
ottogin wants to merge 1 commit intokarpathy:masterfrom
Open
60% more efficient autoresearch via better training analysis#353ottogin wants to merge 1 commit intokarpathy:masterfrom
ottogin wants to merge 1 commit intokarpathy:masterfrom
Conversation
Link to a notable fork
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Hi! While experimenting with autoresearch, I noticed that the agent has very limited observability into the training process and rarely looks beyond the final validation loss.
I updated
train.pyto log more training statistics and added an analysis step where the agent uses Python to inspect training dynamics. This consistently improves BPB.I ran this comparison multiple times—there’s some noise, but extended logging + analysis consistently leads to lower BPB. Experiments were run on H100 with Claude Opus 4.6 via Claude Code.
I think this could be helpful for others working with autoresearch, so in this PR I’m adding a link to my code as a notable fork. I’m also happy to submit a PR with all the changes to the main repo if you think that makes sense.
Details: https://github.com/ottogin/auto-log-research