#FROM vllm/vllm-openai:v0.19.0-cu130
FROM vllm/vllm-openai:cu130-nightly-x86_64

# 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 && \
    pip install --no-cache-dir pandas

# 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