0ce7744324
📁 Files changed: 6 📝 Lines changed: 110 • deploy.md • Dockerfile • Dockerfile.rocm • README.md • deploy.sh • cli.py
117 lines
4.0 KiB
Docker
117 lines
4.0 KiB
Docker
# Tensors + ComfyUI for AMD ROCm (gfx1030 only - slimmed)
|
|
# Multi-stage build, strips unused GPU architectures
|
|
|
|
# ============================================================================
|
|
# STAGE 1: Builder - install everything
|
|
# ============================================================================
|
|
FROM saiden/sd-rocm AS builder
|
|
|
|
WORKDIR /workspace
|
|
|
|
# Install build deps
|
|
RUN apt-get update && apt-get install -y --no-install-recommends git curl && \
|
|
rm -rf /var/lib/apt/lists/*
|
|
|
|
# Install ComfyUI (skip torch - use base image's ROCm torch)
|
|
# Filter out torch ecosystem, install deps without pulling CUDA
|
|
RUN git clone --depth 1 https://github.com/comfyanonymous/ComfyUI.git && \
|
|
cd ComfyUI && \
|
|
grep -v "^torch" requirements.txt | \
|
|
grep -v "^torchvision" | \
|
|
grep -v "^torchaudio" | \
|
|
grep -v "^torchsde" | \
|
|
grep -v "^nvidia" | \
|
|
grep -v "^triton" > requirements-no-torch.txt && \
|
|
pip install --break-system-packages --no-cache-dir -r requirements-no-torch.txt && \
|
|
pip install --break-system-packages --no-cache-dir --no-deps torchsde && \
|
|
pip install --break-system-packages --no-cache-dir trampoline scipy && \
|
|
# Remove any nvidia packages that snuck in
|
|
pip uninstall -y nvidia-cublas-cu12 nvidia-cuda-runtime-cu12 nvidia-cudnn-cu12 \
|
|
nvidia-cufft-cu12 nvidia-curand-cu12 nvidia-cusolver-cu12 nvidia-cusparse-cu12 \
|
|
nvidia-nccl-cu12 nvidia-nvjitlink-cu12 nvidia-nvtx-cu12 nvidia-nvshmem-cu12 \
|
|
triton 2>/dev/null || true && \
|
|
rm -rf .git .github tests* notebooks *.md
|
|
|
|
# Install ComfyUI Manager
|
|
RUN cd /workspace/ComfyUI/custom_nodes && \
|
|
git clone --depth 1 https://github.com/ltdrdata/ComfyUI-Manager.git && \
|
|
cd ComfyUI-Manager && \
|
|
pip install --break-system-packages --no-cache-dir -r requirements.txt && \
|
|
rm -rf .git .github tests *.md
|
|
|
|
# Install tensors
|
|
COPY . /tmp/tensors
|
|
RUN pip install --break-system-packages --no-cache-dir /tmp/tensors'[server]' && \
|
|
rm -rf /tmp/tensors
|
|
|
|
# ============================================================================
|
|
# STAGE 2: Runtime - slim ROCm for gfx1030 only
|
|
# ============================================================================
|
|
FROM saiden/sd-rocm
|
|
|
|
WORKDIR /workspace
|
|
|
|
# Strip unused GPU architectures (keep only gfx1030)
|
|
# hipblaslt has NO gfx1030 support - remove entirely (saves 11GB)
|
|
# rocblas - keep only gfx1030 kernels (saves 3.5GB)
|
|
RUN rm -rf /opt/rocm*/lib/hipblaslt/library/* && \
|
|
find /opt/rocm*/lib/rocblas/library/ -type f \
|
|
! -name '*gfx1030*' \
|
|
! -name 'TensileLibrary.dat' \
|
|
! -name 'TensileLibrary_lazy_gfx1030.dat' \
|
|
-delete && \
|
|
rm -rf /opt/rocm*/share/doc /opt/rocm*/share/html
|
|
|
|
# Install only runtime deps
|
|
RUN apt-get update && apt-get install -y --no-install-recommends curl && \
|
|
rm -rf /var/lib/apt/lists/* && \
|
|
apt-get clean
|
|
|
|
# Copy Python packages from builder
|
|
COPY --from=builder /usr/local/lib/python3.12/dist-packages /usr/local/lib/python3.12/dist-packages
|
|
COPY --from=builder /usr/local/bin/tsr /usr/local/bin/tsr
|
|
|
|
# Copy ComfyUI
|
|
COPY --from=builder /workspace/ComfyUI /workspace/ComfyUI
|
|
|
|
# Configure tensors
|
|
RUN mkdir -p /root/.config/tensors && \
|
|
cat > /root/.config/tensors/config.toml << 'EOF'
|
|
[paths]
|
|
models_dir = "/workspace/ComfyUI/models"
|
|
checkpoints = "/workspace/ComfyUI/models/checkpoints"
|
|
loras = "/workspace/ComfyUI/models/loras"
|
|
vae = "/workspace/ComfyUI/models/vae"
|
|
embeddings = "/workspace/ComfyUI/models/embeddings"
|
|
|
|
[comfyui]
|
|
url = "http://127.0.0.1:8188"
|
|
EOF
|
|
|
|
# Startup script
|
|
RUN cat > /workspace/start.sh << 'EOF'
|
|
#!/bin/bash
|
|
export HSA_OVERRIDE_GFX_VERSION=10.3.0
|
|
|
|
# Start ComfyUI in background (internal only)
|
|
python3 /workspace/ComfyUI/main.py --listen 127.0.0.1 --port 8188 &
|
|
|
|
# Wait for ComfyUI
|
|
echo "Waiting for ComfyUI..."
|
|
until curl -s http://127.0.0.1:8188/system_stats > /dev/null 2>&1; do
|
|
sleep 1
|
|
done
|
|
echo "ComfyUI ready"
|
|
|
|
# Start tensors API (exposed)
|
|
exec tsr serve --host 0.0.0.0 --port 51200
|
|
EOF
|
|
RUN chmod +x /workspace/start.sh
|
|
|
|
ENV HSA_OVERRIDE_GFX_VERSION=10.3.0
|
|
|
|
EXPOSE 51200
|
|
|
|
ENTRYPOINT []
|
|
CMD ["/bin/bash", "/workspace/start.sh"]
|