Billy D.
a16ffff73f
- Add Dockerfiles for nvidia, rdna2, strixhalo, and intel GPU targets - Add ray-serve modules (embeddings, whisper, tts, llm, reranker) - Add Gitea Actions workflow for automated builds - Add Makefile for local development - Update README with comprehensive documentation
73 lines
2.3 KiB
Docker
73 lines
2.3 KiB
Docker
# Ray Worker for AMD Strix Halo (gfx1151 / RDNA 3.5)
|
|
# Pre-bakes all dependencies for fast startup
|
|
#
|
|
# Build from llm-workflows root:
|
|
# docker build -t git.daviestechlabs.io/daviestechlabs/ray-worker-strixhalo:latest -f dockerfiles/Dockerfile.ray-worker-strixhalo .
|
|
#
|
|
# Multi-stage build to ensure Python 3.11.11 matches Ray head node
|
|
|
|
# Stage 1: Extract ROCm 7.1 libraries from vendor image
|
|
FROM docker.io/rocm/pytorch:rocm7.1_ubuntu24.04_py3.12_pytorch_release_2.9.1 AS rocm-libs
|
|
|
|
# Stage 2: Build on Ray base with Python 3.11
|
|
FROM rayproject/ray:2.53.0-py311 AS base
|
|
|
|
# Copy ROCm stack from vendor image
|
|
COPY --from=rocm-libs /opt/rocm /opt/rocm
|
|
|
|
# Set up ROCm environment
|
|
ENV ROCM_HOME=/opt/rocm
|
|
ENV PATH="${ROCM_HOME}/bin:${ROCM_HOME}/llvm/bin:${PATH}"
|
|
ENV LD_LIBRARY_PATH="${ROCM_HOME}/lib:${ROCM_HOME}/lib64:${LD_LIBRARY_PATH}"
|
|
ENV HSA_PATH="${ROCM_HOME}/hsa"
|
|
ENV HIP_PATH="${ROCM_HOME}/hip"
|
|
|
|
# ROCm environment for AMD Strix Halo (gfx1151 / RDNA 3.5)
|
|
ENV HIP_VISIBLE_DEVICES=0
|
|
ENV HSA_ENABLE_SDMA=0
|
|
ENV PYTORCH_HIP_ALLOC_CONF=expandable_segments:True,max_split_size_mb:512
|
|
ENV HSA_OVERRIDE_GFX_VERSION=11.0.0
|
|
ENV ROCM_TARGET_LST=gfx1151,gfx1100
|
|
ENV PYTHONPATH=/app
|
|
|
|
WORKDIR /app
|
|
|
|
# Install ROCm system dependencies
|
|
USER root
|
|
RUN apt-get update && apt-get install -y --no-install-recommends \
|
|
libelf1 \
|
|
libnuma1 \
|
|
libdrm2 \
|
|
libdrm-amdgpu1 \
|
|
kmod \
|
|
&& rm -rf /var/lib/apt/lists/*
|
|
USER ray
|
|
|
|
# WORKAROUND: ROCm/ROCm#5853 - Standard PyTorch ROCm wheels cause segfault
|
|
# in libhsa-runtime64.so during VRAM allocation on gfx1151 (Strix Halo).
|
|
# TheRock gfx110X-all packages provide Python 3.11 compatible wheels.
|
|
RUN pip install --no-cache-dir \
|
|
--index-url https://rocm.nightlies.amd.com/v2/gfx110X-all/ \
|
|
torch torchaudio torchvision
|
|
|
|
# Install Ray Serve and AI inference dependencies
|
|
RUN pip install --no-cache-dir \
|
|
vllm \
|
|
transformers \
|
|
accelerate \
|
|
sentence-transformers \
|
|
httpx \
|
|
numpy \
|
|
scipy
|
|
|
|
# Pre-download common models for faster cold starts
|
|
RUN python3 -c "from sentence_transformers import SentenceTransformer; SentenceTransformer('BAAI/bge-large-en-v1.5')" || true
|
|
|
|
# Copy Ray Serve Python code
|
|
COPY ray-serve/ /app/ray_serve/
|
|
|
|
# Ray worker entrypoint
|
|
COPY --chmod=755 dockerfiles/ray-entrypoint.sh /app/ray-entrypoint.sh
|
|
|
|
ENTRYPOINT ["/app/ray-entrypoint.sh"]
|