Build for AArch64¶

# Add user to qaic group to access Cloud AI devices without root
sudo usermod -aG qaic $USER
newgrp qaic

# Create a python virtual environment
python3.10 -m venv qaic-vllm-venv

source qaic-vllm-venv/bin/activate

# Install the current release version of `Qualcomm Efficient-Transformers <https://github.com/quic/efficient-transformers>`_ (vLLM with qaic support requires efficient-transformers for model exporting and compilation)
pip install -U pip

pip install git+https://github.com/quic/efficient-transformers@release/v1.21.0

# Build Triton 3.0.0 dependency for AArch64
mkdir triton-3.0.0
cd triton-3.0.0
git clone https://github.com/triton-lang/triton.git -b release/3.0.x --depth=1
cd ..

# LLVM git cloning

mkdir llvm-main && cd llvm-main
export LLVM_HASH="$(cat /path_to/triton-3.0.0/triton/cmake/llvm-hash.txt)"

git clone --recursive https://github.com/llvm/llvm-project.git -b main
cd llvm-project && git checkout ${LLVM_HASH}

# LLVM building:
pip install ninja cmake
mkdir llvm-project/build; cd llvm-project/build;
cmake -G Ninja -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=ON ../llvm -DLLVM_ENABLE_PROJECTS="mlir;llvm" -DLLVM_TARGETS_TO_BUILD="host;NVPTX;AMDGPU"; ninja;


# Triton building:
cd /path_to/triton-3.0.0/triton/python
export LLVM_BUILD_DIR=/path_to/llvm-main/llvm-project/build
export LLVM_INCLUDE_DIRS=$LLVM_BUILD_DIR/include
export LLVM_LIBRARY_DIR=$LLVM_BUILD_DIR/lib
export LLVM_SYSPATH=$LLVM_BUILD_DIR
pip wheel .

pip install triton-3.0.0-cp310-cp310-linux_aarch64.whl

cd /back/to/root/path

# Clone the vLLM repo, and apply the patch for qaic backend support
git clone https://github.com/vllm-project/vllm.git

cd vllm

git checkout v0.10.1

git apply /opt/qti-aic/integrations/vllm/qaic_vllm.patch

# Set environment variables and install
export VLLM_TARGET_DEVICE="qaic"

pip install -e .