disable fused lm head chunking by default#1904
Merged
Conversation
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
| ), | ||
| ), | ||
| ] = "auto" | ||
| ] = "disabled" |
There was a problem hiding this comment.
Missing changelog for config default change
Low Severity
The default for model.fused_lm_head_chunk_size changed in src/prime_rl/configs/trainer.py, but there is no corresponding CHANGELOG.md entry for this config behavior change. This makes the default change harder to discover for users relying on documented config migrations.
Triggered by project rule: BugBot Instructions
65536 OOMs on A6000 without fused lm head chunking. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
The benchmark baselines were generated with chunking enabled, so the regression test must explicitly enable it regardless of the default. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This test uses seq_len=65536 which needs chunking to avoid OOM on A6000 GPUs. 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
fused_lm_head_chunk_sizedefault from"auto"to"disabled"🤖 Generated with Claude Code
Note
Medium Risk
Changes a model-level default (
fused_lm_head_chunk_size) that affects training execution and performance/memory characteristics across runs. CI/integration tests now pin an explicit chunk size, but other consumers may see behavior changes if relying on the previous default.Overview
Disables fused LM head chunking by default by changing
ModelConfig.fused_lm_head_chunk_sizefrom"auto"to"disabled".Bench/CI flows are updated to explicitly opt in when needed:
run_single_benchmark.pyaccepts/forwards--model.fused-lm-head-chunk-size, the RL multi-run integrationtrainer.tomlpinsfused_lm_head_chunk_size = 8192, and benchmark regression tests pass--fused-lm-head-chunk-size 8192to keep results stable.Written by Cursor Bugbot for commit fc7621d. This will update automatically on new commits. Configure here.