45 lines
1.9 KiB
Docker
45 lines
1.9 KiB
Docker
#FROM vllm/vllm-openai:v0.19.0-cu130
|
|
FROM vllm/vllm-openai:nightly
|
|
#FROM vllm/vllm-openai:glm51-cu130
|
|
|
|
# Install LMCache for KV cache offloading / sharing across nodes
|
|
# Build with system CUDA 13.0 for Blackwell (B200)
|
|
RUN apt-get update && apt-get install -y git \
|
|
libcusolver-dev-13-0 \
|
|
libcusparse-dev-13-0 \
|
|
libcublas-dev-13-0 \
|
|
libcurand-dev-13-0 \
|
|
libcufft-dev-13-0 \
|
|
libnvjitlink-dev-13-0 && \
|
|
git clone https://github.com/biondizzle/LMCache.git /tmp/lmcache && \
|
|
cd /tmp/lmcache && \
|
|
git checkout feat/redis-ttl && \
|
|
CUDA_HOME=/usr/local/cuda \
|
|
TORCH_CUDA_ARCH_LIST="10.0" \
|
|
pip install --no-cache-dir --no-build-isolation . && \
|
|
rm -rf /tmp/lmcache && export CACHE_BUSTER=2
|
|
|
|
# Copy over nemotron reasonong parser
|
|
#COPY ./super_v3_reasoning_parser.py /opt/super_v3_reasoning_parser.py
|
|
|
|
# Copy over deepseek tool call parser with MTP fixes
|
|
#COPY deepseekv32_tool_parser.py /usr/local/lib/python3.12/dist-packages/vllm/tool_parsers/deepseekv32_tool_parser.py
|
|
|
|
# Copy over minimax tool call parser with kwargs fixes
|
|
#COPY minimax_tool_parser.py /usr/local/lib/python3.12/dist-packages/vllm/tool_parsers/minimax_tool_parser.py
|
|
|
|
# Copy over minimax parsers with kwargs fixes
|
|
#COPY minimax_tool_parser.py /usr/local/lib/python3.12/dist-packages/vllm/tool_parsers/minimax_tool_parser.py
|
|
#COPY minimax_m2_parser.py /usr/local/lib/python3.12/dist-packages/vllm/parser/minimax_m2_parser.py
|
|
|
|
|
|
# Patch tool parser for GLM regex fix
|
|
#COPY glm4_moe_tool_parser.py /usr/local/lib/python3.12/dist-packages/vllm/tool_parsers/glm4_moe_tool_parser.py
|
|
#COPY utils.py /usr/local/lib/python3.12/dist-packages/vllm/tool_parsers/utils.py
|
|
|
|
# Patch hf renderer to force string content format for GLM models
|
|
# This fixes the issue where tool response content is dropped
|
|
#COPY hf.py /usr/local/lib/python3.12/dist-packages/vllm/renderers/hf.py
|
|
|
|
# Make sure we have the latest up to date chat template
|
|
COPY glm_5.1_chat_template.jinja /opt/chat_template.jinja |