nick/comfyui-serverless
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Limit SageAttention to A100/H100 due to cross-compile issues
Some checks failed
Build and Push Docker Image / build (push) Has been cancelled
Details

Browse Source 
The sm90 kernels use wgmma instructions that can't be compiled for
sm86/sm89 targets. Restricting to 8.0 (A100) and 9.0 (H100) only.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This commit is contained in:
Debian
2026-01-03 21:57:09 +00:00
parent 352128aa39
commit 9f71e6db57
1 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
Dockerfile
View File

@@ -66,9 +66,9 @@ ENV EXT_PARALLEL=2
ENV NVCC_APPEND_FLAGS="--threads 2" ENV NVCC_APPEND_FLAGS="--threads 2"
ENV MAX_JOBS=4 ENV MAX_JOBS=4
# Target RunPod GPU architectures: # Target RunPod GPU architectures:
# 8.0 = A100, 8.6 = A10/RTX 3090, 8.9 = RTX 4090/L40, 9.0 = H100/H200 # 8.0 = A100, 9.0 = H100/H200
# Note: Blackwell (10.0) not yet supported by SageAttention # Note: 8.6/8.9 excluded due to SageAttention sm90 kernel cross-compile issues
ENV TORCH_CUDA_ARCH_LIST="8.0;8.6;8.9;9.0" ENV TORCH_CUDA_ARCH_LIST="8.0;9.0"
RUN git clone https://github.com/thu-ml/SageAttention.git && \ RUN git clone https://github.com/thu-ml/SageAttention.git && \
cd SageAttention && \ cd SageAttention && \
git checkout 2aecfa8 && \ git checkout 2aecfa8 && \