#!/bin/bash

read -p "Enter config (0 for fine-tuned, 1 for base): " CONFIG
export CONFIG=${CONFIG:-0}

if [ $CONFIG -eq 0 ]; then
    export CUDA_VISIBLE_DEVICES=1
    export MODEL_NAME="/home/data2/sltian/code/evaluation_agent_dev/LLaMA-Factory/saves/qwen2.5-3b/eval-agent-vbench-base-table" #"ea-dev/eval-agent_qwen2.5-3b_instruct_ckpt471_base"
    export PORT=12333
    export GPU_MEMORY_UTILIZATION=0.7
else
    export CUDA_VISIBLE_DEVICES=2
    export MODEL_NAME="qwen/Qwen2.5-3B-Instruct"
    export PORT=12334
    export GPU_MEMORY_UTILIZATION=0.7
fi

# Launch fine-tuned Qwen2.5-3B eval agent model server

echo "Starting fine-tuned Qwen2.5-3B eval agent server on 0.0.0.0:${PORT}..."
echo "Model: ${MODEL_NAME}"
echo "GPU Memory Utilization: ${GPU_MEMORY_UTILIZATION}"

python -m vllm.entrypoints.openai.api_server \
    --model ${MODEL_NAME} \
    --host 0.0.0.0 \
    --port ${PORT} \
    --gpu-memory-utilization ${GPU_MEMORY_UTILIZATION} \
    --trust-remote-code true \
    --max-model-len 16384 \
    --served-model-name eval-agent