entrypoint.sh

#!/bin/bash

mkdir -p data

HF_MODEL_PATH=Qwen/Qwen2.5-1.5B-Instruct
HF_MODEL_NAME=$(basename ${HF_MODEL_PATH})
LOCAL_MODEL_PATH=./data/${HF_MODEL_NAME}

# TODO: use your own key and put into secret
VLLM_SERVE_API_KEY=demo-123

# download model
HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download ${HF_MODEL_PATH} --local-dir ${LOCAL_MODEL_PATH}

# start vllm server
vllm serve ${LOCAL_MODEL_PATH} --served-model-name ${HF_MODEL_NAME} --api-key ${VLLM_SERVE_API_KEY} &

# start frontend
cd /app/playground && npm run dev &

# start backend
export OPENAI_API_BASE=http://127.0.0.1:8000/v1
export OPENAI_PROXY_URL=
export OPENAI_MODEL=${HF_MODEL_NAME}
export OPENAI_API_KEY=${VLLM_SERVE_API_KEY}
cd /app && task run