一键部署各种 internlm

deploy_internlm.sh

cd ~

available_models=(internlm/internlm-chat-7b internlm/internlm-chat-7b-8k internlm/internlm-chat-20b lmdeploy/turbomind-internlm-chat-20b-w4)
num_gpus=$(nvidia-smi -L | wc -l)
model_name=$(dirname $1)

found=false
for model in "${available_models[@]}"
do
  if [ "$model" == "$1" ]
  then
    found=true
    break
  fi
done

if [ "$found" == false ]
then
  echo "Usage: $0 \$model_name"
  echo "Available model_name is: ${available_models[@]}" 
  exit 1
fi

if pip show lmdeploy >/dev/null 2>&1
then
  echo "Package lmdeploy is already installed"
else
  echo "Installing Python dependencies..."
  pip install lmdeploy socksio
fi

if git lfs >/dev/null 2>&1
then
  echo "Git lfs is already installed"
else
  echo "Installing git lfs extension..."
  curl -s https://packagecloud.io/install/repositories/github/git-lfs/script.deb.sh | sudo bash
  sudo apt-get install -y git-lfs
  git lfs install
fi

echo "Download repo and weights..."
if [ ! -d ${model} ]
then
  git clone https://huggingface.co/${model} ${model}
fi

# if model start with turbomind
if [[ $model == lmdeploy/* ]]
then
  python3 -m lmdeploy.serve.gradio.app --tp=${num_gpus} --server_name=0.0.0.0 ${model}
else
  echo "Converting model to turbomind..."
  python3 -m lmdeploy.serve.turbomind.deploy --tp=${num_gpus} --dst_path=${model}_workspace ${model_name} ${model}

  echo "Launching"
  python3 -m lmdeploy.serve.gradio.app --tp=${num_gpus} --server_name=0.0.0.0 ${model}_workspace
fi