blazerye / DrugAssist

DrugAssist: A Large Language Model for Molecule Optimization
https://arxiv.org/abs/2401.10334
124 stars 10 forks source link

MacOSX M1 installation failure #1

Open lucatoldo opened 4 months ago

lucatoldo commented 4 months ago

Dear Authors, on a MacOSX M1 running Sonoma, I did the following:

curl https://repo.anaconda.com/miniconda/Miniconda3-latest-MacOSX-arm64.sh -o ~/miniconda3/miniconda.sh
bash ~/miniconda3/miniconda.sh -b -u -p ~/miniconda3
rm -rf ~/miniconda3/miniconda.sh
~/miniconda3/bin/conda init bash
~/miniconda3/bin/conda init zsh
git clone https://github.com/blazerye/DrugAssist.git\ncd DrugAssist
cd DrugAssist
conda create -n drugassist python=3.8 -y
conda activate drugassist
pip install -r requirements.txt

The last step failed due to missing packages: "packaging" "torch".

After manually pip installing them, sh run_sft_lora.sh failed due to missing packages:

Unfortunately the latter could not be installed due to dependency with a non available package (cpuinfo). Would be great if you could fix.

blazerye commented 3 months ago

Our model is trained on a Linux server with 8 NVIDIA Tesla A100-SXM4-40GB GPUs, and we strongly recommend using a server with NVIDIA GPUs for training, as training on Mac may be very slow. If you do need to train on Mac, please remove the vllm and flash-attn packages from the requirements.txt file, while keeping the other packages and their corresponding versions unchanged.