biondizzle/deepseek-v4-quant
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: add missing staging_kernel.py to Dockerfile — BF16→E2M1+UE4M3 quantization was never in container

Browse Source
This commit is contained in:
biondizzle
2026-05-12 21:21:24 +00:00
parent 91d7d9bad7
commit d88ea9842b
1 changed files with 2 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
Dockerfile
View File

@@ -22,6 +22,8 @@ RUN cd /root/DeepGEMM && python3 setup.py build_ext --inplace && PATCH_CACHE_BUS
ARG PATCH_CACHE_BUSTER=69
# Copy our DeepSeek V4 patch over vLLM's model file
COPY patches/deepseek_v4.py /usr/local/lib/python3.12/dist-packages/vllm/model_executor/models/deepseek_v4.py
# Copy the NVFP4 staging kernel (BF16→E2M1+UE4M3 quantization for activations)
COPY patches/staging_kernel.py /usr/local/lib/python3.12/dist-packages/vllm/model_executor/models/staging_kernel.py
# Verify everything imports
RUN python3 -c "import deep_gemm; print('DeepGEMM NVFP4 OK')" && \