[rollout,vllm] Fix DP args and local_rank for Ray NOSET_VISIBLE_DEVICES #5233
+52
−2
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Outdated suggestions cannot be applied.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
data_parallel_size > 1(not only when EP is enabled).RAY_EXPERIMENTAL_NOSET_CUDA_VISIBLE_DEVICES=1anddata_parallel_size > 1,compute the correct vLLM-local
LOCAL_RANK(local rank within TP×PP) to avoidDP adjusted local rank ... out of boundsand incorrect GPU binding.
Context / Motivation
In DP configs (typically
dp=2,tp=1,pp=1,nnodes=1):ray.get_runtime_context().get_accelerator_ids()to derive local rank can produce a rank that becomesout-of-bounds after vLLM DP adjustment.
Checklist