NVIDIA-NeMo / NVIDIA-NeMo/RL

[INSTALL-8] deep-ep hardcodes CUDA 12 cccl path; fails on CUDA 13

Open
#2,712 1 comment 0 reactions 1 assignee Claimed by @kajalj22 View on GitHub
bug Documentation p1
Dominant language
Python
Stars
2k
Forks
561
Avg merge
4d 5h
Merged PRs (30d)
145

Description

## Problem

`deep-ep` expects `cccl` at the CUDA 12 path. In CUDA 13, `cccl` was moved to a subdirectory, so `cuda/std/tuple` now lives under `cccl/`. The build fails unless the updated path is manually added to `CPATH`. This is undocumented.

## Steps to reproduce

Attempt bare-metal installation on a CUDA 13 system. The `deep-ep` build fails with a path error related to `cccl`.

## Suggested fix

Update `deep-ep`'s build configuration to detect the CUDA version and resolve the correct `cccl` path accordingly. Document the manual `CPATH` workaround in the meantime.

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.