0.9.0.1

2025-06-04 03:22:03 +00:00
parent 3ea7d34e83
commit 3d115911aa
4 changed files with 69 additions and 47 deletions
--- a/lmcache/Dockerfile
+++ b/lmcache/Dockerfile
@@ -0,0 +1,56 @@
+ARG CUDA_VERSION=12.8.1
+ARG IMAGE_DISTRO=ubuntu24.04
+ARG PYTHON_VERSION=3.12
+
+# ---------- Builder Base ----------
+FROM nvcr.io/nvidia/cuda:${CUDA_VERSION}-devel-${IMAGE_DISTRO} AS base
+
+ARG TORCH_CUDA_ARCH_LIST="9.0a"
+ENV TORCH_CUDA_ARCH_LIST=${TORCH_CUDA_ARCH_LIST}
+ENV DEBIAN_FRONTEND=noninteractive
+
+# Install dependencies
+RUN apt update && apt upgrade -y && \
+    apt install -y --no-install-recommends \
+        curl \
+        git \
+        libibverbs-dev \
+        zlib1g-dev && \
+    apt clean && \
+    rm -rf /var/lib/apt/lists/* /var/cache/apt/archives
+
+# Set compiler paths
+ENV CC=/usr/bin/gcc
+ENV CXX=/usr/bin/g++
+
+# Install uv
+RUN curl -LsSf https://astral.sh/uv/install.sh | env UV_INSTALL_DIR=/usr/local/bin sh
+
+# Set up workspace and virtualenv
+WORKDIR /workspace
+ARG PYTHON_VERSION
+RUN uv venv -p ${PYTHON_VERSION} --seed --python-preference only-managed
+
+# Activate uv venv
+ENV VIRTUAL_ENV=/workspace/.venv
+ENV PATH=${VIRTUAL_ENV}/bin:${PATH}
+ENV CUDA_HOME=/usr/local/cuda
+ENV LD_LIBRARY_PATH=${CUDA_HOME}/lib64:${LD_LIBRARY_PATH}
+
+# Install Python deps in venv
+RUN uv pip install numpy==2.0.0
+RUN uv pip install torch==2.7.0 torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu128
+
+FROM base AS build-base
+
+# Install build tools and dependencies
+RUN uv pip install -U build cmake ninja pybind11 setuptools==79.0.1 wheel
+
+# Clone and build LMCache wheel
+RUN git clone https://github.com/rajesh-s/LMCache.git -b v0.3.0 && \
+    cd LMCache && \
+    uv pip install setuptools_scm && \
+    python -m build --wheel --no-isolation && \
+    cp dist/*.whl /workspace/
+
+CMD ["/bin/bash"]
--- a/lmcache/lmcache-0.3.1.dev1-cp312-cp312-linux_aarch64.whl
+++ b/lmcache/lmcache-0.3.1.dev1-cp312-cp312-linux_aarch64.whl
--- a/vllm/Dockerfile
+++ b/vllm/Dockerfile
@@ -48,7 +48,7 @@ ENV LD_LIBRARY_PATH=${CUDA_HOME}/lib64:${LD_LIBRARY_PATH}

 RUN uv pip install numpy==2.0.0
 # Install pytorch nightly
-RUN uv pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu128
+RUN uv pip install torch==2.7.0+cu128 torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu128

 FROM base AS build-base
 RUN mkdir /wheels
@@ -68,6 +68,7 @@ RUN cd triton && \
    git submodule update --init --recursive -j 8 && \
    uv build python --wheel --no-build-isolation -o /wheels

+RUN export MAX_JOBS=10
 FROM build-base AS build-xformers
 ARG XFORMERS_REF=v0.0.30
 ARG XFORMERS_BUILD_VERSION=0.0.30+cu128
@@ -125,54 +126,19 @@ RUN uv clean
 # python3-config https://github.com/astral-sh/uv/issues/10263
 RUN export PATH="$(dirname $(realpath .venv/bin/python)):$PATH"

+# Install build tools and dependencies
+RUN uv pip install -U build cmake ninja pybind11 setuptools==79.0.1 wheel

-# LMCache should be auto-integrated in v1
-# InfiniStore dependencies -> not needed with patched LMCache below
-# RUN apt-get update && apt-get install -y --no-install-recommends \
-#     libuv1-dev \
-#     libflatbuffers-dev \
-#     libspdlog-dev \
-#     libfmt-dev \
-#     ibverbs-utils \
-#     libibverbs-dev \
-#     libboost-dev \
-#     libboost-stacktrace-dev \
-#     git \
-#     curl \
-#     build-essential
-
-# RUN git clone https://github.com/bytedance/InfiniStore.git && \
-#     cd InfiniStore && git checkout 0.2.33 && \
-#     pip install -e .
-# RUN apt update
-# RUN apt install -y cmake libuv1-dev libspdlog-dev libboost-dev libboost-all-dev # pybind11-dev python3-dev
-# RUN uv pip install pybind11
-# RUN git clone -b v1.12.0 https://github.com/google/flatbuffers.git && \
-#   cd flatbuffers && \
-#   cmake -B build -DFLATBUFFERS_BUILD_TESTS=OFF && \
-#   cmake --build build -j && \
-#   cmake --install build && cd ..
-# RUN git clone https://github.com/bytedance/InfiniStore
-# RUN sed -i 's|^INCLUDES = -I/usr/local/.*|INCLUDES = -I/root/.local/share/uv/python/cpython-3.12.10-linux-aarch64-gnu/include/python3.12|' InfiniStore/src/Makefile
-# RUN cd InfiniStore/src && make
-# RUN cd InfiniStore && pip install --no-deps --no-build-isolation -e .
-
-# # # LMCache dependencies
-# RUN uv pip install -U aiofiles pyyaml redis nvtx safetensors transformers psutil aiohttp sortedcontainers prometheus_client msgspec
-# # RUN git clone https://github.com/LMCache/torchac_cuda.git && \
-# #     cd torchac_cuda && \
-# #     python setup.py install
-
-# RUN git clone https://github.com/rajesh-s/LMCache.git && \
-#     cd LMCache && \
-#     sed -i 's/2\.5\.1/2.6.0/g' pyproject.toml setup.py && \
-#     sed 's#numpy==1\.26\.4#numpy#g' pyproject.toml setup.py requirements.txt && \
-#     python setup.py install
+# Clone and build LMCache wheel without Infinistore that is broken on aarch64
+RUN git clone https://github.com/rajesh-s/LMCache.git -b v0.3.0 && \
+    cd LMCache && \
+    uv pip install setuptools_scm && \
+    python -m build --wheel --no-isolation && \
+    cp dist/*.whl /workspace/

 # Enable hf-transfer
 ENV HF_HUB_ENABLE_HF_TRANSFER=1
-RUN uv pip install numpy datasets aiohttp 
-# ==2.0.0
+RUN uv pip install datasets aiohttp

 # Install nsys for profiling
 ARG NSYS_URL=https://developer.nvidia.com/downloads/assets/tools/secure/nsight-systems/2025_3/
--- a/vllm/README.md
+++ b/vllm/README.md
@@ -4,6 +4,6 @@ Hosted [here](https://hub.docker.com/repository/docker/rajesh550/gh200-vllm)

 ```bash
 docker login
-docker build --platform linux/arm64 -t rajesh550/gh200-vllm:0.8.1 .
-docker push rajesh550/gh200-vllm:0.8.1
+docker buildx build --platform linux/arm64 --memory=600g -t rajesh550/gh200-vllm:0.9.0.1 .
+docker push rajesh550/gh200-vllm:0.9.0.1
 ```