[multi-lora] skip pause/resume for lora with merge=false#1677
Merged
Conversation
Contributor
There was a problem hiding this comment.
Code Review
This pull request adds a training script for fully async GRPO on GSM8K with Megatron and LoRA. It also updates the weight synchronization logic to avoid pausing generation when using non-merged LoRA. Feedback indicates that the condition for skipping the pause is too broad and could cause errors on non-Megatron strategies; a more robust implementation using getattr and a backend check for vLLM is recommended.
erictang000
added a commit
that referenced
this pull request
May 16, 2026
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
pausing and resuming generation are not needed when updating lora weights in vLLM - we can simply rely on
load_lora_adapter(https://docs.vllm.ai/en/stable/features/lora/#in-place-lora-reloading) to directly update the weights in the engine without worrying about corrupted weight state.Results for fully async RL with megatron + vllm LoRA - we can see that removing the pause/resume for the merge lora = false case still works and has an identical reward curve!