Fix IgGM container: install cuda-nvcc and set CUDA_HOME - #1
Merged
Conversation
Without a compiler toolkit, torch.utils.cpp_extension.CUDA_HOME stays unset, and deepspeed's op_builder raises MissingCUDAException while importing deepspeed itself (before the code's own `except ImportError` guard around DS4Sci_EvoformerAttention can catch it), crashing design.py on start with: deepspeed.ops.op_builder.builder.MissingCUDAException: CUDA_HOME does not exist Add cuda-nvcc=11.7 (matching the pinned cudatoolkit=11.7.1) and point CUDA_HOME at the conda env so deepspeed's compatibility check succeeds.
|
Bugbot is not enabled for your account, so this pull request was not reviewed. Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs. |
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
docker run --gpus all/apptainer exec --nvcrashes on startup duringimport deepspeed:environment.yamlonly pulls in the CUDA runtime (cudatoolkit/cuda-runtime), not thenvcccompiler, sotorch.utils.cpp_extension.CUDA_HOMEis never set. DeepSpeed checks this unconditionally while importing itself (ingit_version_info.py->op_builder.is_compatible()), before the repo's ownexcept ImportErrorguard aroundfrom deepspeed.ops.deepspeed4science import DS4Sci_EvoformerAttention(IgGM/model/module/attention/gating_multihead_attention.py) gets a chance to catch it — so it is not silently skipped.cuda-nvcc=11.7(matching the pinnedcudatoolkit=11.7.1) alongside the pytorch/pytorch-cuda reinstall step, and setENV CUDA_HOME="${ENV_DIR}"so$CUDA_HOME/bin/nvccresolves inside the conda env.Test plan
2026-05-28_06abc56_nv-cuda11_weightsimage and reproduced the crash on an M3 GPU node (A40) viaapptainer exec --nv ... python /app/design.py ....cuda-nvcc=11.7.99exists in thenvidiaconda channel, matching the pinnedcudatoolkit=11.7.1/cuda-runtime=11.7.1.design.pyruns past the deepspeed import.