mirror of
https://github.com/dogkeeper886/ollama37.git
synced 2025-12-10 07:46:59 +00:00
Sync with upstream ollama/ollama and restore Tesla K80 (compute 3.7) support
This commit represents a complete rework after pulling the latest changes from official ollama/ollama repository and re-applying Tesla K80 compatibility patches. ## Key Changes ### CUDA Compute Capability 3.7 Support (Tesla K80) - Added sm_37 (compute 3.7) to CMAKE_CUDA_ARCHITECTURES in CMakeLists.txt - Updated CMakePresets.json to include compute 3.7 in "CUDA 11" preset - Using 37-virtual (PTX with JIT compilation) for maximum compatibility ### Legacy Toolchain Compatibility - **NVIDIA Driver**: 470.256.02 (last version supporting Kepler/K80) - **CUDA Version**: 11.4.4 (last CUDA 11.x supporting compute 3.7) - **GCC Version**: 10.5.0 (required by CUDA 11.4 host_config.h) ### CPU Architecture Trade-offs Due to GCC 10.5 limitation, sacrificed newer CPU optimizations: - Alderlake CPU variant enabled WITHOUT AVX_VNNI (requires GCC 11+) - Still supports: SSE4.2, AVX, F16C, AVX2, BMI2, FMA - Performance impact: ~3-7% on newer CPUs (acceptable for K80 compatibility) ### Build System Updates - Modified ml/backend/ggml/ggml/src/ggml-cuda/CMakeLists.txt for compute 3.7 - Added -Wno-deprecated-gpu-targets flag to suppress warnings - Updated ml/backend/ggml/ggml/src/CMakeLists.txt for Alderlake without AVX_VNNI ### Upstream Sync Merged latest llama.cpp changes including: - Enhanced KV cache management with ISWA and hybrid memory support - Improved multi-modal support (mtmd framework) - New model architectures (Gemma3, Llama4, Qwen3, etc.) - GPU backend improvements for CUDA, Metal, and ROCm - Updated quantization support and GGUF format handling ### Documentation - Updated CLAUDE.md with comprehensive build instructions - Documented toolchain constraints and CPU architecture trade-offs - Removed outdated CI/CD workflows (tesla-k80-*.yml) - Cleaned up temporary development artifacts ## Rationale This fork maintains Tesla K80 GPU support (compute 3.7) which was dropped in official Ollama due to legacy driver/CUDA requirements. The toolchain constraint creates a deadlock: - K80 → Driver 470 → CUDA 11.4 → GCC 10 → No AVX_VNNI We accept the loss of cutting-edge CPU optimizations to enable running modern LLMs on legacy but still capable Tesla K80 hardware (12GB VRAM per GPU). 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
This commit is contained in:
Shang Chieh Tseng
118
Dockerfile
118
Dockerfile
@@ -1,11 +1,13 @@
|
||||
# vim: filetype=dockerfile
|
||||
|
||||
ARG FLAVOR=${TARGETARCH}
|
||||
ARG PARALLEL=8
|
||||
|
||||
ARG ROCMVERSION=6.3.3
|
||||
ARG JETPACK5VERSION=r35.4.1
|
||||
ARG JETPACK6VERSION=r36.4.0
|
||||
ARG CMAKEVERSION=3.31.2
|
||||
ARG VULKANVERSION=1.4.321.1
|
||||
|
||||
# We require gcc v10 minimum. v10.3 has regressions, so the rockylinux 8.5 AppStream has the latest compatible version
|
||||
FROM --platform=linux/amd64 rocm/dev-almalinux-8:${ROCMVERSION}-complete AS base-amd64
|
||||
@@ -16,6 +18,16 @@ RUN yum install -y yum-utils \
|
||||
&& dnf install -y ccache \
|
||||
&& yum-config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel8/x86_64/cuda-rhel8.repo
|
||||
ENV PATH=/opt/rh/gcc-toolset-10/root/usr/bin:$PATH
|
||||
ARG VULKANVERSION
|
||||
RUN wget https://sdk.lunarg.com/sdk/download/${VULKANVERSION}/linux/vulkansdk-linux-x86_64-${VULKANVERSION}.tar.xz -O /tmp/vulkansdk-linux-x86_64-${VULKANVERSION}.tar.xz \
|
||||
&& tar xvf /tmp/vulkansdk-linux-x86_64-${VULKANVERSION}.tar.xz \
|
||||
&& dnf -y install ninja-build \
|
||||
&& ln -s /usr/bin/python3 /usr/bin/python \
|
||||
&& /${VULKANVERSION}/vulkansdk -j 8 vulkan-headers \
|
||||
&& /${VULKANVERSION}/vulkansdk -j 8 shaderc
|
||||
RUN cp -r /${VULKANVERSION}/x86_64/include/* /usr/local/include/ \
|
||||
&& cp -r /${VULKANVERSION}/x86_64/lib/* /usr/local/lib
|
||||
ENV PATH=/${VULKANVERSION}/x86_64/bin:$PATH
|
||||
|
||||
FROM --platform=linux/arm64 almalinux:8 AS base-arm64
|
||||
# install epel-release for ccache
|
||||
@@ -34,26 +46,52 @@ ENV LDFLAGS=-s
|
||||
FROM base AS cpu
|
||||
RUN dnf install -y gcc-toolset-11-gcc gcc-toolset-11-gcc-c++
|
||||
ENV PATH=/opt/rh/gcc-toolset-11/root/usr/bin:$PATH
|
||||
ARG PARALLEL
|
||||
RUN --mount=type=cache,target=/root/.ccache \
|
||||
cmake --preset 'CPU' \
|
||||
&& cmake --build --parallel --preset 'CPU' \
|
||||
&& cmake --install build --component CPU --strip --parallel 8
|
||||
&& cmake --build --parallel ${PARALLEL} --preset 'CPU' \
|
||||
&& cmake --install build --component CPU --strip --parallel ${PARALLEL}
|
||||
|
||||
FROM base AS cuda-11
|
||||
ARG CUDA11VERSION=11.8
|
||||
RUN dnf install -y cuda-toolkit-${CUDA11VERSION//./-}
|
||||
ENV PATH=/usr/local/cuda-11/bin:$PATH
|
||||
ARG PARALLEL
|
||||
RUN --mount=type=cache,target=/root/.ccache \
|
||||
cmake --preset 'CUDA 11' -DOLLAMA_RUNNER_DIR="cuda_v11" \
|
||||
&& cmake --build --parallel ${PARALLEL} --preset 'CUDA 11' \
|
||||
&& cmake --install build --component CUDA --strip --parallel ${PARALLEL}
|
||||
|
||||
FROM base AS cuda-12
|
||||
ARG CUDA12VERSION=12.8
|
||||
RUN dnf install -y cuda-toolkit-${CUDA12VERSION//./-}
|
||||
ENV PATH=/usr/local/cuda-12/bin:$PATH
|
||||
ARG PARALLEL
|
||||
RUN --mount=type=cache,target=/root/.ccache \
|
||||
cmake --preset 'CUDA 12' \
|
||||
&& cmake --build --parallel --preset 'CUDA 12' \
|
||||
&& cmake --install build --component CUDA --strip --parallel 8
|
||||
cmake --preset 'CUDA 12' -DOLLAMA_RUNNER_DIR="cuda_v12"\
|
||||
&& cmake --build --parallel ${PARALLEL} --preset 'CUDA 12' \
|
||||
&& cmake --install build --component CUDA --strip --parallel ${PARALLEL}
|
||||
|
||||
|
||||
FROM base AS cuda-13
|
||||
ARG CUDA13VERSION=13.0
|
||||
RUN dnf install -y cuda-toolkit-${CUDA13VERSION//./-}
|
||||
ENV PATH=/usr/local/cuda-13/bin:$PATH
|
||||
ARG PARALLEL
|
||||
RUN --mount=type=cache,target=/root/.ccache \
|
||||
cmake --preset 'CUDA 13' -DOLLAMA_RUNNER_DIR="cuda_v13" \
|
||||
&& cmake --build --parallel ${PARALLEL} --preset 'CUDA 13' \
|
||||
&& cmake --install build --component CUDA --strip --parallel ${PARALLEL}
|
||||
|
||||
|
||||
FROM base AS rocm-6
|
||||
ENV PATH=/opt/rocm/hcc/bin:/opt/rocm/hip/bin:/opt/rocm/bin:/opt/rocm/hcc/bin:$PATH
|
||||
ARG PARALLEL
|
||||
RUN --mount=type=cache,target=/root/.ccache \
|
||||
cmake --preset 'ROCm 6' \
|
||||
&& cmake --build --parallel --preset 'ROCm 6' \
|
||||
&& cmake --install build --component HIP --strip --parallel 8
|
||||
cmake --preset 'ROCm 6' -DOLLAMA_RUNNER_DIR="rocm" \
|
||||
&& cmake --build --parallel ${PARALLEL} --preset 'ROCm 6' \
|
||||
&& cmake --install build --component HIP --strip --parallel ${PARALLEL}
|
||||
RUN rm -f dist/lib/ollama/rocm/rocblas/library/*gfx90[06]*
|
||||
|
||||
FROM --platform=linux/arm64 nvcr.io/nvidia/l4t-jetpack:${JETPACK5VERSION} AS jetpack-5
|
||||
ARG CMAKEVERSION
|
||||
@@ -61,10 +99,11 @@ RUN apt-get update && apt-get install -y curl ccache \
|
||||
&& curl -fsSL https://github.com/Kitware/CMake/releases/download/v${CMAKEVERSION}/cmake-${CMAKEVERSION}-linux-$(uname -m).tar.gz | tar xz -C /usr/local --strip-components 1
|
||||
COPY CMakeLists.txt CMakePresets.json .
|
||||
COPY ml/backend/ggml/ggml ml/backend/ggml/ggml
|
||||
ARG PARALLEL
|
||||
RUN --mount=type=cache,target=/root/.ccache \
|
||||
cmake --preset 'JetPack 5' \
|
||||
&& cmake --build --parallel --preset 'JetPack 5' \
|
||||
&& cmake --install build --component CUDA --strip --parallel 8
|
||||
cmake --preset 'JetPack 5' -DOLLAMA_RUNNER_DIR="cuda_jetpack5" \
|
||||
&& cmake --build --parallel ${PARALLEL} --preset 'JetPack 5' \
|
||||
&& cmake --install build --component CUDA --strip --parallel ${PARALLEL}
|
||||
|
||||
FROM --platform=linux/arm64 nvcr.io/nvidia/l4t-jetpack:${JETPACK6VERSION} AS jetpack-6
|
||||
ARG CMAKEVERSION
|
||||
@@ -72,10 +111,18 @@ RUN apt-get update && apt-get install -y curl ccache \
|
||||
&& curl -fsSL https://github.com/Kitware/CMake/releases/download/v${CMAKEVERSION}/cmake-${CMAKEVERSION}-linux-$(uname -m).tar.gz | tar xz -C /usr/local --strip-components 1
|
||||
COPY CMakeLists.txt CMakePresets.json .
|
||||
COPY ml/backend/ggml/ggml ml/backend/ggml/ggml
|
||||
ARG PARALLEL
|
||||
RUN --mount=type=cache,target=/root/.ccache \
|
||||
cmake --preset 'JetPack 6' \
|
||||
&& cmake --build --parallel --preset 'JetPack 6' \
|
||||
&& cmake --install build --component CUDA --strip --parallel 8
|
||||
cmake --preset 'JetPack 6' -DOLLAMA_RUNNER_DIR="cuda_jetpack6" \
|
||||
&& cmake --build --parallel ${PARALLEL} --preset 'JetPack 6' \
|
||||
&& cmake --install build --component CUDA --strip --parallel ${PARALLEL}
|
||||
|
||||
FROM base AS vulkan
|
||||
RUN --mount=type=cache,target=/root/.ccache \
|
||||
cmake --preset 'Vulkan' -DOLLAMA_RUNNER_DIR="vulkan" \
|
||||
&& cmake --build --parallel --preset 'Vulkan' \
|
||||
&& cmake --install build --component Vulkan --strip --parallel 8
|
||||
|
||||
|
||||
FROM base AS build
|
||||
WORKDIR /go/src/github.com/ollama/ollama
|
||||
@@ -86,29 +133,62 @@ RUN go mod download
|
||||
COPY . .
|
||||
ARG GOFLAGS="'-ldflags=-w -s'"
|
||||
ENV CGO_ENABLED=1
|
||||
ARG CGO_CFLAGS
|
||||
ARG CGO_CXXFLAGS
|
||||
RUN --mount=type=cache,target=/root/.cache/go-build \
|
||||
go build -trimpath -buildmode=pie -o /bin/ollama .
|
||||
|
||||
FROM --platform=linux/amd64 scratch AS amd64
|
||||
COPY --from=cuda-12 dist/lib/ollama /lib/ollama
|
||||
# COPY --from=cuda-11 dist/lib/ollama/ /lib/ollama/
|
||||
COPY --from=cuda-12 dist/lib/ollama /lib/ollama/
|
||||
COPY --from=cuda-13 dist/lib/ollama /lib/ollama/
|
||||
COPY --from=vulkan dist/lib/ollama /lib/ollama/
|
||||
|
||||
FROM --platform=linux/arm64 scratch AS arm64
|
||||
COPY --from=cuda-12 dist/lib/ollama /lib/ollama/cuda_sbsa
|
||||
COPY --from=jetpack-5 dist/lib/ollama /lib/ollama/cuda_jetpack5
|
||||
COPY --from=jetpack-6 dist/lib/ollama /lib/ollama/cuda_jetpack6
|
||||
# COPY --from=cuda-11 dist/lib/ollama/ /lib/ollama/
|
||||
COPY --from=cuda-12 dist/lib/ollama /lib/ollama/
|
||||
COPY --from=cuda-13 dist/lib/ollama/ /lib/ollama/
|
||||
COPY --from=jetpack-5 dist/lib/ollama/ /lib/ollama/
|
||||
COPY --from=jetpack-6 dist/lib/ollama/ /lib/ollama/
|
||||
|
||||
FROM scratch AS rocm
|
||||
COPY --from=rocm-6 dist/lib/ollama /lib/ollama
|
||||
|
||||
FROM ${FLAVOR} AS archive
|
||||
ARG VULKANVERSION
|
||||
COPY --from=cpu dist/lib/ollama /lib/ollama
|
||||
COPY --from=build /bin/ollama /bin/ollama
|
||||
|
||||
FROM ubuntu:24.04
|
||||
# Temporary opt-out stages for Vulkan
|
||||
FROM --platform=linux/amd64 scratch AS amd64_novulkan
|
||||
# COPY --from=cuda-11 dist/lib/ollama/ /lib/ollama/
|
||||
COPY --from=cuda-12 dist/lib/ollama /lib/ollama/
|
||||
COPY --from=cuda-13 dist/lib/ollama /lib/ollama/
|
||||
FROM arm64 AS arm64_novulkan
|
||||
FROM ${FLAVOR}_novulkan AS archive_novulkan
|
||||
COPY --from=cpu dist/lib/ollama /lib/ollama
|
||||
COPY --from=build /bin/ollama /bin/ollama
|
||||
FROM ubuntu:24.04 AS novulkan
|
||||
RUN apt-get update \
|
||||
&& apt-get install -y ca-certificates \
|
||||
&& apt-get clean \
|
||||
&& rm -rf /var/lib/apt/lists/*
|
||||
COPY --from=archive_novulkan /bin /usr/bin
|
||||
ENV PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
|
||||
COPY --from=archive_novulkan /lib/ollama /usr/lib/ollama
|
||||
ENV LD_LIBRARY_PATH=/usr/local/nvidia/lib:/usr/local/nvidia/lib64
|
||||
ENV NVIDIA_DRIVER_CAPABILITIES=compute,utility
|
||||
ENV NVIDIA_VISIBLE_DEVICES=all
|
||||
ENV OLLAMA_HOST=0.0.0.0:11434
|
||||
EXPOSE 11434
|
||||
ENTRYPOINT ["/bin/ollama"]
|
||||
CMD ["serve"]
|
||||
|
||||
FROM ubuntu:24.04 AS default
|
||||
RUN apt-get update \
|
||||
&& apt-get install -y ca-certificates libvulkan1 \
|
||||
&& apt-get clean \
|
||||
&& rm -rf /var/lib/apt/lists/*
|
||||
COPY --from=archive /bin /usr/bin
|
||||
ENV PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
|
||||
COPY --from=archive /lib/ollama /usr/lib/ollama
|
||||
|
||||
Reference in New Issue
Block a user