89 lines
3.0 KiB
Docker
89 lines
3.0 KiB
Docker
FROM nvidia/cuda:13.0.2-cudnn-devel-ubuntu24.04
|
|
|
|
# Set non-interactive frontend to prevent apt prompts
|
|
ENV DEBIAN_FRONTEND=noninteractive
|
|
|
|
# CRITICAL: Allow pip to install globally on Ubuntu 24.04 without a venv
|
|
ENV PIP_BREAK_SYSTEM_PACKAGES=1
|
|
|
|
# Set the base directory environment variable
|
|
ENV VLLM_BASE_DIR=/workspace/vllm
|
|
|
|
# 1. Install System Dependencies
|
|
# Added 'git', 'wget', and 'python3-pip' as they are required for the script steps
|
|
RUN apt-get update && apt-get install -y \
|
|
curl \
|
|
cmake \
|
|
build-essential \
|
|
ninja-build \
|
|
python3-dev \
|
|
python3-pip \
|
|
git \
|
|
wget \
|
|
&& rm -rf /var/lib/apt/lists/*
|
|
|
|
# Setup Workspace
|
|
WORKDIR $VLLM_BASE_DIR
|
|
|
|
# 2. Download Tiktoken files
|
|
RUN mkdir -p tiktoken_encodings && \
|
|
wget -O tiktoken_encodings/o200k_base.tiktoken "https://openaipublic.blob.core.windows.net/encodings/o200k_base.tiktoken" && \
|
|
wget -O tiktoken_encodings/cl100k_base.tiktoken "https://openaipublic.blob.core.windows.net/encodings/cl100k_base.tiktoken"
|
|
|
|
# 3. Set Environment Variables
|
|
# Note: TORCH_CUDA_ARCH_LIST=12.1a is very specific (Hopper/H100 usually).
|
|
# Ensure this matches your target hardware.
|
|
ENV TORCH_CUDA_ARCH_LIST=12.1a
|
|
ENV TRITON_PTXAS_PATH=/usr/local/cuda/bin/ptxas
|
|
ENV TIKTOKEN_ENCODINGS_BASE=$VLLM_BASE_DIR/tiktoken_encodings
|
|
|
|
# --- CACHE BUSTER ---
|
|
# Change this argument to force a re-download of PyTorch/FlashInfer
|
|
ARG CACHEBUST_DEPS=1
|
|
|
|
# 4. Install Python Dependencies (Using pip instead of uv)
|
|
|
|
# Install PyTorch for CUDA 13.0
|
|
RUN pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu130
|
|
|
|
# Install Helper libraries
|
|
RUN pip install xgrammar triton
|
|
|
|
# Install FlashInfer
|
|
# Note: Using the same index URLs as provided in your script
|
|
RUN pip install flashinfer-python --no-deps --index-url https://flashinfer.ai/whl && \
|
|
pip install flashinfer-cubin --index-url https://flashinfer.ai/whl && \
|
|
pip install flashinfer-jit-cache --index-url https://flashinfer.ai/whl/cu130
|
|
|
|
# Install fast safetensors to improve loading speeds
|
|
RUN pip install fastsafetensors>=0.1.10
|
|
|
|
# --- VLLM SOURCE CACHE BUSTER ---
|
|
# Change THIS argument to force a fresh git clone and rebuild of vLLM
|
|
# without re-installing the dependencies above.
|
|
ARG CACHEBUST_VLLM=1
|
|
|
|
# 5. Clone and Build vLLM
|
|
RUN git clone --recursive https://github.com/vllm-project/vllm.git
|
|
WORKDIR $VLLM_BASE_DIR/vllm
|
|
|
|
# Prepare build requirements
|
|
RUN python3 use_existing_torch.py && \
|
|
sed -i "/flashinfer/d" requirements/cuda.txt && \
|
|
pip install -r requirements/build.txt
|
|
|
|
# TEMPORARY - apply NVFP4 patch
|
|
RUN curl -L https://patch-diff.githubusercontent.com/raw/vllm-project/vllm/pull/29242.diff | git apply
|
|
|
|
# Final Build
|
|
# Uses --no-build-isolation to respect the pre-installed Torch/FlashInfer
|
|
# Changed -e (editable) to . (standard install) for better Docker portability
|
|
RUN pip install --no-build-isolation . -v
|
|
|
|
# Set the final workdir
|
|
WORKDIR $VLLM_BASE_DIR
|
|
|
|
# Copy clustering script
|
|
COPY run-cluster-node.sh $VLLM_BASE_DIR/
|
|
RUN chmod +x $VLLM_BASE_DIR/run-cluster-node.sh
|