Note: please specify PATH_LOCAL and PATH_DOCKER according to your preference.
sudo docker run -it —gpus all \
—name FlashLLM \
—privileged \
-v LOCAL_PATH:DOCKER_PATH \
nvcr.io/nvidia/pytorch:22.07-py3 bash
git clone https://github.com/AlibabaResearch/flash-llm.git
cd Flash-LLM
git submodule update --init --recursive
source Init_FlashLLM.sh
cd $FlashLLM_HOME/third_party/FasterTransformer && git am ../ft.patch
cd $FlashLLM_HOME/third_party/sputnik && git am ../sputnik.patch
The libSpMM_API.so and SpMM_API.cuh will be available for easy integration after:
cd $FlashLLM_HOME/build && make -j