rag基座已经搭建好

llm_server/start_asr.sh

@@ -1,62 +1,44 @@
 #!/usr/bin/env bash
 set -euo pipefail
 
-# 配置区域（可根据实际情况修改）
-MODEL_BASE_PATH="${MODEL_BASE_PATH:-/mnt/Qwen3-Omni-30B-A3B-Instruct-NVFP4}"
-THINKER_PATH="${MODEL_BASE_PATH}/thinker"
-CONTAINER_NAME="${CONTAINER_NAME:-vllm-thinker}"
-HOST_PORT="${HOST_PORT:-8091}"
-GPU_DEVICES="${GPU_DEVICES:-all}"          # 例如 "0,1" 或 "all"
-TENSOR_PARALLEL_SIZE="${TENSOR_PARALLEL_SIZE:-1}"
-MAX_MODEL_LEN="${MAX_MODEL_LEN:-32768}"
-VLLM_IMAGE="${VLLM_IMAGE:-vllm/vllm-openai:latest}"
+# 自动获取脚本所在目录
+SCRIPT_DIR="$(cd -- "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+VENV_DIR="${SCRIPT_DIR}/.venv_audio"
+MODEL_DIR="${SCRIPT_DIR}/modelscope/models/Qwen/Qwen3-ASR-1.7B"
+HOST_PORT="${HOST_PORT:-8901}"
 
-# 检查模型路径是否存在
-if [[ ! -d "$THINKER_PATH" ]]; then
-  echo "错误: 模型路径不存在: $THINKER_PATH"
-  echo "请设置环境变量 MODEL_BASE_PATH 指向正确的模型根目录（包含 thinker 子目录）"
+# 激活虚拟环境
+source "${VENV_DIR}/bin/activate"
+
+# 检查端口占用
+if lsof -i "tcp:${HOST_PORT}" -t >/dev/null 2>&1; then
+  echo "Port ${HOST_PORT} already in use" >&2
   exit 1
 fi
 
-# 检查 Docker 可用性
-if command -v docker >/dev/null 2>&1 && docker info >/dev/null 2>&1; then
-  DOCKER_CMD="docker"
-else
-  DOCKER_CMD="sudo docker"
-fi
+echo "Starting vLLM server on port ${HOST_PORT} ..."
+echo "Model path: ${MODEL_DIR}"
 
-# 停止并移除可能已存在的同名容器
-${DOCKER_CMD} rm -f "${CONTAINER_NAME}" >/dev/null 2>&1 || true
+# 进入脚本目录
+cd "${SCRIPT_DIR}"
 
-echo "启动 Thinker 服务..."
-echo "模型路径: $THINKER_PATH"
-echo "服务端口: $HOST_PORT"
-echo "Tensor Parallel Size: $TENSOR_PARALLEL_SIZE"
+# 设置环境变量
+export VLLM_USE_MODELSCOPE=true
+export MODELSCOPE_DOWNLOAD_MODE=local
+export MODELSCOPE_HUB_ENABLE_FILE_CHECK=false
+export MODELSCOPE_CACHE="${SCRIPT_DIR}/modelscope"
+export HF_HUB_OFFLINE=1
+export TRANSFORMERS_OFFLINE=1
 
-# 启动容器
-${DOCKER_CMD} run -d \
-  --name "${CONTAINER_NAME}" \
-  --gpus "${GPU_DEVICES}" \
-  --ipc=host \
-  --network host \
-  -v "${THINKER_PATH}:/model:ro" \
-  -e VLLM_USE_MODELSCOPE=false \
-  "${VLLM_IMAGE}" \
-  sh -c "pip install --no-cache-dir nvidia-modelopt && \
-         vllm serve /model \
-           --port ${HOST_PORT} \
-           --quantization modelopt_fp4 \
-           --dtype bfloat16 \
-           --max-model-len ${MAX_MODEL_LEN} \
-           --tensor-parallel-size ${TENSOR_PARALLEL_SIZE} \
-           --trust-remote-code \
-           --enforce-eager"
-
-# 等待服务启动
-sleep 3
-
-# 健康检查
-echo "健康检查 (等待几秒后):"
-echo "  curl http://127.0.0.1:${HOST_PORT}/health"
-echo "查看日志:"
-echo "  ${DOCKER_CMD} logs -f ${CONTAINER_NAME}"
+# 启动 vLLM 服务
+vllm serve "${MODEL_DIR}" \
+  --port "${HOST_PORT}" \
+  --host 0.0.0.0 \
+  --quantization fp8 \
+  --dtype auto \
+  --kv-cache-dtype fp8 \
+  --max-model-len 16384 \
+  --allowed-local-media-path / \
+  --tensor-parallel-size 1 \
+  --gpu-memory-utilization 0.10 \
+  --max-num-batched-tokens 4096