Installation with ROCm
Aphrodite supports AMD GPUs using ROCm 6.1.
Requirements
- Linux (or WSL on Windows)
- Python 3.8 - 3.11
- GPU: MI200 (gfx90a), MI300 (gfx942), RX 7900 Series (gfx1100)
- ROCm 6.1
Installation with Docker
You can build Aphrodite Engine from source. First, build a docker image from the provided Dockerfile.rocm
, then launch a container from the image.
To build Aphrodite on high-end datacenter GPUs (e.g. MI300X), run this:
sh
DOCKER_BUILDKIT=1 docker build -f Dockerfile.rocm -t aphrodite-rocm .
To build Aphrodite on NAVI GPUs (e.g. RTX 7900 XTX), run this:
sh
DOCKER_BUILDKIT=1 docker build --build-arg BUILD_FA="0" -f Dockerfile.rocm aphrodite-rocm .
Then run your image:
sh
docker run -it \
--network=host \
--group-add=video \
--ipc=host \
--cap-add=SYS_PTRACE \
--security-opt seccomp=unconfined \
--device /dev/kfd \
--device /dev/dri \
-v ~/.cache/huggingface/root/.cache/huggingface \
aphrodite-rocm \
bash
Installation from source
You can also build Aphrodite from source, but it's more complicated, so we recommend Docker.
You will need the following installed beforehand:
Then install Triton for ROCm. You may also Install CK Flash Attention if needed.
WARNING
You may need to downgrade ninja
version to 1.10.
Finally, build Aphrodite:
sh
git clone https://github.com/PygmalionAI/aphrodite-engine.git
cd aphrodite-engine
chmod +x ./amdpatch.sh
./amdpatch.sh
pip install -U -r requirements-rocm.txt
python setup.py develop # pip install -e . won't work for now