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

bump cache buster 32 for packed FP4 mxf4nvf4 fix

Browse Source
This commit is contained in:
biondizzle
2026-05-11 21:59:56 +00:00
parent c85b84b0fe
commit 79d866995f
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
Dockerfile
View File

@@ -8,7 +8,7 @@ RUN apt-get update && apt-get install -y git screen cmake && rm -rf /var/lib/apt
# Clone and build DeepGEMM with NVFP4 mega_moe kernel
# CACHE_BUSTER: increment to force fresh clone
RUN git clone -b nvfp4-mega-moe https://sweetapi.com/biondizzle/DeepGEMM.git /root/DeepGEMM && CACHE_BUSTER=31
RUN git clone -b nvfp4-mega-moe https://sweetapi.com/biondizzle/DeepGEMM.git /root/DeepGEMM && CACHE_BUSTER=32
# Build DeepGEMM with proper CUDA/NVRTC paths
ENV CPATH="/usr/local/lib/python3.12/dist-packages/flashinfer/data/cutlass/include:/usr/local/lib/python3.12/dist-packages/nvidia/cu13/include:/usr/local/cuda-13.0/include:${CPATH}"
@@ -19,7 +19,7 @@ RUN ln -sf /usr/local/lib/python3.12/dist-packages/nvidia/cu13/lib/libnvrtc.so.1
RUN cd /root/DeepGEMM && python3 setup.py build_ext --inplace
# Bust cache for patch changes — ARG before COPY ensures layer invalidation
ARG PATCH_CACHE_BUSTER=31
ARG PATCH_CACHE_BUSTER=32
# 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