fix: distribute code_gen to Ray workers via working_dir#1000
Draft
fix: distribute code_gen to Ray workers via working_dir#1000
Conversation
…_executable py_executable pointed into the eval container's venv which doesn't exist on vLLM deployment worker nodes, causing all check_correctness_remote tasks to fall back to the head node with no distribution. Replace with Ray working_dir (Gym root) + pip install from distributed source so workers in the deployment container get lcb_integration and nemo_gym without requiring a pre-installed Gym venv. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
cmunley1
requested changes
Apr 17, 2026
| "**/.venv", | ||
| "**/venv", | ||
| "docs/", | ||
| "resources/*.png", |
Contributor
There was a problem hiding this comment.
should this exclude also *jsonl, cache, results, maybe other test related things?
| "docs/", | ||
| "resources/*.png", | ||
| ], | ||
| "pip": [".[dev]"], |
Contributor
There was a problem hiding this comment.
probably doesnt need dev right?
Contributor
There was a problem hiding this comment.
could adding a pip install here be slower (on first task i guess)? since before it just use sys.executable?
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The idea
We try to allow passing Gym's contents to head ray's object store
Summary
py_executable=sys.executableincheck_correctness_remote'sruntime_envwithworking_dirpointing to the Gym rootpy_executablepointed into the eval container's venv — an absolute path that doesn't exist on vLLM deployment worker nodes, causing all Ray tasks to fall back to the head node with no distributionworking_dirhas Ray zip and distribute the Gym source to all workers;pip: [".[dev]"]installs nemo-gym from that distributed source (no internet needed);PYTHONPATHkeepslcb_integrationimportable (no pyproject.toml)