Merge pull request #35 from MooreThreads/update_vllm_demo_script

Update vllm demo script

Author: gliangMT

vllm/demo/model_url.json

@@ -0,0 +1,4 @@
+{
+    "DeepSeek-R1-Distill-Qwen-1.5B": "https://www.modelscope.cn/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B.git",
+    "Qwen2.5-0.5B-Instruct": "https://www.modelscope.cn/Qwen/Qwen2.5-0.5B-Instruct.git"
+}

vllm/demo/run_server_ds_r1_dstl_qw1.5b.sh

@@ -0,0 +1,80 @@
+#!/bin/bash
+set -e
+
+MODEL_NAME="$1"
+CONFIG_FILE="model_config.json"
+
+if [ -z "$MODEL_NAME" ]; then
+    echo "× Please provide the model name, for example:"
+    echo "   ./run_vllm_serving.sh DeepSeek-R1-Distill-Qwen-1.5B"
+    exit 1
+fi
+
+# 用 Python 解析 JSON 获取 URL
+MODEL_URL=$(python3 -c "
+import json
+config_file = '$CONFIG_FILE'
+model_name = '$MODEL_NAME'
+with open(config_file, 'r') as f:
+    data = json.load(f)
+print(data.get(model_name, ''))
+")
+
+if [ -z "$MODEL_URL" ]; then
+    echo "× $MODEL_NAME is not supported yet, please refer to the website to try other models: https://docs.mthreads.com/mtt/mtt-doc-online/compability"
+    exit 1
+fi
+
+echo "√ 找到模型 URL: $MODEL_URL"
+
+# 目录和日志路径
+CURRENT_DIR=$(pwd)
+MODEL_DIR="/data/mtt/models"
+CONVERTED_MODEL_DIR="/data/mtt/models_convert"
+LOG_FILE="/data/mtt/logs/model_server.log"
+MODEL_CHECK_FILE="$MODEL_DIR/$MODEL_NAME/model.safetensors"  
+SUCCESS_MESSAGE="INFO:     Started server process"
+
+# 确保目录存在
+mkdir -p "$MODEL_DIR" "$CONVERTED_MODEL_DIR" "$(dirname "$LOG_FILE")"
+
+# 检查模型是否已经存在
+if [ -f "$MODEL_CHECK_FILE" ]; then
+    echo "√ The model file already exists. Skip the download step."
+else
+    echo "⬇ The model file does not exist, start downloading the model..."
+    cd "$MODEL_DIR"
+    apt update && apt install -y git-lfs jq
+    git lfs install
+    git clone "$MODEL_URL" "$MODEL_NAME"
+    echo "√ Model download completed."
+fi
+
+# 权重转换
+cd "${CURRENT_DIR}/.."
+./convert_weight.sh "$MODEL_DIR/$MODEL_NAME" 1
+
+# 启动 vLLM 服务器
+python -m vllm.entrypoints.openai.api_server \
+        --model "$CONVERTED_MODEL_DIR/$MODEL_NAME-tp1-convert" \
+        --trust-remote-code \
+        --tensor-parallel-size 1 \
+        -pp 1 \
+        --block-size 64 \
+        --max-model-len 2048 \
+        --disable-log-stats \
+        --disable-log-requests \
+        --device "musa" \
+        --served-model-name model-develop_test > "$LOG_FILE" 2>&1 &
+
+pid=$!
+echo "Wait for the service to start..."
+while true; do
+    if grep -q "$SUCCESS_MESSAGE" "$LOG_FILE"; then
+        echo "√ Service has been started. If it does not work, check the log: $LOG_FILE"
+        break
+    else
+        echo "Wait for the service to start..."
+        sleep 5  # 每隔 5 秒检查日志文件
+    fi
+done