会议记录

configs/model_config.py

@@ -34,32 +34,24 @@ EMBEDDING_DEVICE = "cuda" if torch.cuda.is_available() else "mps" if torch.backe
 
 llm_model_dict = {
     "chatglm-6b": {
-        "name": "chatglm-6b",  # "name"修改为fastchat服务中的"model_name"
+        "local_model_path": "THUDM/chatglm-6b",
-        "pretrained_model_name": "chatglm-6b",
-        "local_model_path": "",
         "api_base_url": "http://localhost:8888/v1",  # "name"修改为fastchat服务中的"api_base_url"
         "api_key": "EMPTY"
     },
 
     "chatglm-6b-int4": {
-        "name": "chatglm-6b-int4",  # "name"修改为fastchat服务中的"model_name"
+        "local_model_path": "THUDM/chatglm-6b-int4",
-        "pretrained_model_name": "chatglm-6b-int4",
-        "local_model_path": "",
         "api_base_url": "http://localhost:8001/v1",  # "name"修改为fastchat服务中的"api_base_url"
         "api_key": "EMPTY"
     },
 
     "chatglm2-6b": {
-        "name": "chatglm2-6b",  # "name"修改为fastchat服务中的"model_name"
+        "local_model_path": "THUDM/chatglm2-6b",
-        "pretrained_model_name": "chatglm2-6b",
-        "local_model_path": "",
         "api_base_url": "http://localhost:8888/v1",  # "name"修改为fastchat服务中的"api_base_url"
         "api_key": "EMPTY"
     },
 
     "vicuna-13b-hf": {
-        "name": "vicuna-13b-hf",  # "name"修改为fastchat服务中的"model_name"
-        "pretrained_model_name": "vicuna-13b-hf",
         "local_model_path": "",
         "api_base_url": "http://localhost:8000/v1",  # "name"修改为fastchat服务中的"api_base_url"
         "api_key": "EMPTY"
@@ -76,11 +68,9 @@ llm_model_dict = {
     # Failed to establish a new connection: [WinError 10060]
     # 则是因为内地和香港的IP都被OPENAI封了，需要切换为日本、新加坡等地
     "openai-chatgpt-3.5": {
-        "name": "gpt-3.5-turbo",
+        "local_model_path": "gpt-3.5-turbo",
-        "pretrained_model_name": "gpt-3.5-turbo",
-        "local_model_path": "",
         "api_base_url": "https://api.openapi.com/v1",
-        "api_key": ""
+        "api_key": os.environ["OPENAI_API_KEY"]
     },
 }
 
@@ -92,6 +82,8 @@ LLM_DEVICE = "cuda" if torch.cuda.is_available() else "mps" if torch.backends.mp
 
 # 日志存储路径
 LOG_PATH = os.path.join(os.path.dirname(os.path.dirname(__file__)), "logs")
+if not os.path.exists(LOG_PATH):
+    os.mkdir(LOG_PATH)
 
 # 知识库默认存储路径
 KB_ROOT_PATH = os.path.join(os.path.dirname(os.path.dirname(__file__)), "knowledge_base")

server/llm_api.py

@@ -11,29 +11,42 @@ def execute_command(command):
 
 
 host_ip = "0.0.0.0"
-port = 8888
+port = 8887
+# 1. llm_model_dict精简；
 
+# 2. 不同任务的日志还是分开；
+
+# 3. 在此处定义args，可选fastchat为每个服务所提供的命令行参数，model_path除外；
+
+# 4. 用logger.removeHandler把它添加的handler删掉，添加我们自己的handler;
+
+# 5. 用watchdog监控第二步的执行情况；
+
+# 6. requirements指定fastchat版本号。
+
+print(llm_model_dict[LLM_MODEL])
 model_path = llm_model_dict[LLM_MODEL]["local_model_path"]
 if not model_path:
     logger.error("local_model_path 不能为空")
 else:
     # 启动任务
-    command1 = f'nohup python -m fastchat.serve.controller >> {LOG_PATH}/fastchat_log.txt 2>&1 &'
+    command1 = f'nohup python -m fastchat.serve.controller >> {LOG_PATH}/controller_log.txt 2>&1 &'
     process1 = execute_command(command1)
     logger.info(f"已执行 {command1}")
     logger.info(f"Process 1 started with PID: {process1}")
 
-    command2 = f'nohup python -m fastchat.serve.model_worker --model-path "{model_path}" --device mps >> {LOG_PATH}/fastchat_log.txt 2>&1 &'
+    command2 = f'nohup python -m fastchat.serve.model_worker --model-path "{model_path}" --device cuda >> {LOG_PATH}/worker_log.txt 2>&1 &'
     process2 = execute_command(command2)
     logger.info(f"已执行 {command2}")
     logger.info(f"Process 2 started with PID: {process2}")
 
-    command3 = f'nohup python -m fastchat.serve.openai_api_server --host "{host_ip}" --port {port} >> {LOG_PATH}/fastchat_log.txt 2>&1 &'
+    command3 = f'nohup python -m fastchat.serve.openai_api_server --host "{host_ip}" --port {port} >> {LOG_PATH}/api_log.txt 2>&1 &'
     process3 = execute_command(command3)
     logger.info(f"已执行 {command3}")
     logger.info(f"Process 3 started with PID: {process3}")
 
-    # TODO: model_worker.log 与 controller.log 存储位置未指定为 LOG_PATH
+    # TODO: model_worker.log 与 controller.log 存储位置未指定为 LOG_PATH --> （hzg0601）model_worker.py,controller.py自行指定的文件写入路径，
+    # TODO(hzg0601): -->而且是写死的，如果想修改路径必须修改fastchat的代码
     logger.info(f"如需查看 llm_api 日志，请前往 {LOG_PATH}")
 
 # 服务启动后接口调用示例：

setup_server.sh

@@ -0,0 +1,30 @@
+[ -d "./logs/" ] && echo "log dir exists" || mkdir "./logs/"
+# controller
+nohup python3 -m fastchat.serve.controller >./logs/controller.log 2>&1 &
+while [ `grep -c "Uvicorn running on" ./logs/controller.log` -eq '0' ];do
+        sleep 1s;
+        echo "wait controller running"
+done
+echo "controller running"
+
+# worker
+nohup python3 -m fastchat.serve.model_worker \
+--model-name 'chatglm2-6b' \
+--model-path THUDM/chatglm2-6b \
+--num-gpus 2 \
+>> ./logs/worker.log 2>&1 &
+
+while [ `grep -c "Uvicorn running on" ./logs/worker.log` -eq '0' ];do
+        sleep 3s;
+        echo "wait worker running"
+done
+echo "worker running"
+
+# webui
+nohup python3 -m fastchat.serve.openai_api_server >> "./logs/openai_server.log" 2>&1 &
+
+while [ `grep -c "Uvicorn running on" ./logs/openai_server.log` -eq '0' ];do
+        sleep 3s;
+        echo "wait openai_server running"
+done
+echo "openai_server running"

shutdown_server.sh

@@ -0,0 +1 @@
+ps -eo user,pid,cmd|grep fastchat|grep -v grep|awk '{print $2}'|xargs kill -9