NVIDIA-NeMo / NVIDIA-NeMo/RL

Add sm_103 (Grace Blackwell) arch support to sglang dependency configuration

Open
#2,144 0 comments 0 reactions 0 assignees View on GitHub
Dominant language
Python
Stars
2k
Forks
561
Avg merge
4d 5h
Merged PRs (30d)
145

Description

## Summary

PR #2134 adds `10.3` (sm_103 / Grace Blackwell / B300) to `TORCH_CUDA_ARCH_LIST` in `docker/Dockerfile`, but the same architecture support needs to be propagated to other places in the repo.

Specifically, the sglang dependencies in [`pyproject.toml` (line 219)](https://github.com/NVIDIA-NeMo/RL/blob/9feb4b0f13e399cbcee3e7150724deaec44f280e/pyproject.toml#L219) and any related sglang configuration should also be updated to ensure sm_103 compatibility when running on B300 hardware.

## Context

- PR #2134 added Blackwell/Grace Blackwell container support in the Dockerfile
- The sglang optional dependency group (and potentially other places) may need similar arch-aware updates to fully support B300

## Related

- #2134

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.