Skip to content

Conversation

@CyCle1024
Copy link
Collaborator

Motivation

Current docker image would fail ci_test due to DeepGemm import failure:

>>> import deep_gemm
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/local/lib/python3.12/dist-packages/deep_gemm/__init__.py", line 16, in <module>
    from . import deep_gemm_cpp  # noqa: F401  # Registers ops into torch.ops without touching CUDA
    ^^^^^^^^^^^^^^^^^^^^^^^^^^^
ImportError: /usr/local/lib/python3.12/dist-packages/deep_gemm/deep_gemm_cpp.abi3.so: undefined symbol: _ZNK3c106SymInt22maybe_as_int_slow_pathEv
>>>

The latest version of sgl-kernel would hack the components of package deep_gemm with a prebuilt dynamic library, this is not compatitable with our no-deps installation of sglang.

Key changes

  1. fix sgl-kernel==0.3.14.post1 in Dockerfile
  2. Reorder dlblas position in pip installation of Dockerfile
@CyCle1024 CyCle1024 requested a review from YanhuiDua December 16, 2025 06:28
@CyCle1024 CyCle1024 merged commit d769950 into InternLM:main Dec 16, 2025
3 of 4 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

2 participants