| ARG BASE_IMAGE=vllm/vllm-openai:gemma4-cu130 |
| FROM ${BASE_IMAGE} |
|
|
| ARG VLLM_REPO=https://github.com/bluecopa/vllm-spectral.git |
| ARG VLLM_BRANCH=spectral-codebook-docker |
| ARG VLLM_COMMIT=008dd7f87fb9de185e536ad30b4d524024ed9b9f |
| ARG HF_REPO_ID=satya007/gemmacut-spectral |
| ARG SIDECAR_SHA256=e47a36c13467cbedf720e7f782b976df3dcda2d989c727113a8315008661a3e4 |
| ARG INCLUDE_SIDECAR=1 |
|
|
| LABEL org.opencontainers.image.title="gemmacut-spectral" |
| LABEL org.opencontainers.image.description="GemmaCut SpectralQuant Phase 2 + Eagle3 vLLM runtime; model weights are not baked into the image." |
| LABEL org.opencontainers.image.source="https://github.com/bluecopa/vllm-spectral" |
| LABEL org.opencontainers.image.revision="${VLLM_COMMIT}" |
|
|
| ENV VLLM_SOURCE=/opt/vllm-spectral \ |
| GEMMACUT_HOME=/opt/gemmacut \ |
| SPECTRAL_SIDECAR=/opt/gemmacut/artifacts/spectral_sidecar_chat_v2.pt \ |
| HF_HUB_DISABLE_XET=1 \ |
| SPECTRAL_TRITON_COMPRESS=1 \ |
| SPECTRAL_TRITON_DEQUANT=1 \ |
| SPECTRAL_CUDA_GRAPH=1 \ |
| SPECTRAL_VERIFY=0 \ |
| DISABLE_HYBRID_KV_CACHE_MANAGER=0 |
|
|
| SHELL ["/bin/bash", "-o", "pipefail", "-c"] |
|
|
| RUN apt-get update && \ |
| apt-get install -y --no-install-recommends \ |
| ca-certificates \ |
| cmake \ |
| git \ |
| ninja-build && \ |
| rm -rf /var/lib/apt/lists/* |
|
|
| RUN git clone --branch "${VLLM_BRANCH}" "${VLLM_REPO}" "${VLLM_SOURCE}" && \ |
| git -C "${VLLM_SOURCE}" checkout "${VLLM_COMMIT}" && \ |
| git -C "${VLLM_SOURCE}" log --oneline -1 |
|
|
| COPY docker/download_sidecar.py /tmp/download_sidecar.py |
| RUN mkdir -p "${GEMMACUT_HOME}/artifacts" && \ |
| if [[ "${INCLUDE_SIDECAR}" == "1" ]]; then \ |
| HF_REPO_ID="${HF_REPO_ID}" \ |
| SIDECAR_SHA256="${SIDECAR_SHA256}" \ |
| python3 /tmp/download_sidecar.py; \ |
| else \ |
| echo "INCLUDE_SIDECAR=0; mount or set SPECTRAL_SIDECAR at runtime"; \ |
| fi && \ |
| rm -f /tmp/download_sidecar.py |
|
|
| COPY docker/entrypoint.sh /usr/local/bin/gemmacut-spectral |
| RUN chmod +x /usr/local/bin/gemmacut-spectral |
|
|
| EXPOSE 8000 |
| ENTRYPOINT ["gemmacut-spectral"] |
| CMD ["serve"] |
|
|
