Merge pull request #2647 from zRzRzRzRzRzRzR/dev

更新即将废弃的启动内容

server/knowledge_base/utils.py

@@ -262,6 +262,10 @@ def make_text_splitter(
         text_splitter_module = importlib.import_module('langchain.text_splitter')
         TextSplitter = getattr(text_splitter_module, "RecursiveCharacterTextSplitter")
         text_splitter = TextSplitter(chunk_size=chunk_size, chunk_overlap=chunk_overlap)
+        
+    # If you use SpacyTextSplitter you can use GPU to do split likes Issue #1287
+    # text_splitter._tokenizer.max_length = 37016792
+    # text_splitter._tokenizer.prefer_gpu()
     return text_splitter
 
 

startup.py

@@ -6,9 +6,8 @@ import sys
 from multiprocessing import Process
 from datetime import datetime
 from pprint import pprint
+from langchain_core._api import deprecated
 
-
-# 设置numexpr最大线程数，默认为CPU核心数
 try:
     import numexpr
 
@@ -33,15 +32,18 @@ from configs import (
     HTTPX_DEFAULT_TIMEOUT,
 )
 from server.utils import (fschat_controller_address, fschat_model_worker_address,
-                          fschat_openai_api_address, set_httpx_config, get_httpx_client,
+                          fschat_openai_api_address, get_httpx_client, get_model_worker_config,
-                          get_model_worker_config, get_all_model_worker_configs,
                           MakeFastAPIOffline, FastAPI, llm_device, embedding_device)
 from server.knowledge_base.migrate import create_tables
 import argparse
-from typing import Tuple, List, Dict
+from typing import List, Dict
 from configs import VERSION
 
 
+@deprecated(
+    since="0.3.0",
+    message="模型启动功能将于 Langchain-Chatchat 0.3.x重写,支持更多模式和加速启动，0.2.x中相关功能将废弃",
+    removal="0.3.0")
 def create_controller_app(
         dispatch_method: str,
         log_level: str = "INFO",
@@ -88,7 +90,7 @@ def create_model_worker_app(log_level: str = "INFO", **kwargs) -> FastAPI:
 
     for k, v in kwargs.items():
         setattr(args, k, v)
-    if worker_class := kwargs.get("langchain_model"): #Langchian支持的模型不用做操作
+    if worker_class := kwargs.get("langchain_model"):  # Langchian支持的模型不用做操作
         from fastchat.serve.base_model_worker import app
         worker = ""
     # 在线模型API
@@ -107,12 +109,12 @@ def create_model_worker_app(log_level: str = "INFO", **kwargs) -> FastAPI:
             import fastchat.serve.vllm_worker
             from fastchat.serve.vllm_worker import VLLMWorker, app, worker_id
             from vllm import AsyncLLMEngine
-            from vllm.engine.arg_utils import AsyncEngineArgs,EngineArgs
+            from vllm.engine.arg_utils import AsyncEngineArgs
 
-            args.tokenizer = args.model_path # 如果tokenizer与model_path不一致在此处添加
+            args.tokenizer = args.model_path
             args.tokenizer_mode = 'auto'
-            args.trust_remote_code= True
+            args.trust_remote_code = True
-            args.download_dir= None
+            args.download_dir = None
             args.load_format = 'auto'
             args.dtype = 'auto'
             args.seed = 0
@@ -154,15 +156,15 @@ def create_model_worker_app(log_level: str = "INFO", **kwargs) -> FastAPI:
             engine = AsyncLLMEngine.from_engine_args(engine_args)
 
             worker = VLLMWorker(
-                        controller_addr = args.controller_address,
+                controller_addr=args.controller_address,
-                        worker_addr = args.worker_address,
+                worker_addr=args.worker_address,
-                        worker_id = worker_id,
+                worker_id=worker_id,
-                        model_path = args.model_path,
+                model_path=args.model_path,
-                        model_names = args.model_names,
+                model_names=args.model_names,
-                        limit_worker_concurrency = args.limit_worker_concurrency,
+                limit_worker_concurrency=args.limit_worker_concurrency,
-                        no_register = args.no_register,
+                no_register=args.no_register,
-                        llm_engine =  engine,
+                llm_engine=engine,
-                        conv_template = args.conv_template,
+                conv_template=args.conv_template,
             )
             sys.modules["fastchat.serve.vllm_worker"].engine = engine
             sys.modules["fastchat.serve.vllm_worker"].worker = worker
@@ -605,8 +607,10 @@ async def start_main_server():
         Python 3.9 has `signal.strsignal(signalnum)` so this closure would not be needed.
         Also, 3.8 includes `signal.valid_signals()` that can be used to create a mapping for the same purpose.
         """
+
         def f(signal_received, frame):
             raise KeyboardInterrupt(f"{signalname} received")
+
         return f
 
     # This will be inherited by the child process if it is forked (not spawned)
@@ -742,12 +746,12 @@ async def start_main_server():
     else:
         try:
             # 保证任务收到SIGINT后，能够正常退出
-            if p:= processes.get("controller"):
+            if p := processes.get("controller"):
                 p.start()
                 p.name = f"{p.name} ({p.pid})"
                 controller_started.wait()  # 等待controller启动完成
 
-            if p:= processes.get("openai_api"):
+            if p := processes.get("openai_api"):
                 p.start()
                 p.name = f"{p.name} ({p.pid})"
 
@@ -763,12 +767,12 @@ async def start_main_server():
             for e in model_worker_started:
                 e.wait()
 
-            if p:= processes.get("api"):
+            if p := processes.get("api"):
                 p.start()
                 p.name = f"{p.name} ({p.pid})"
                 api_started.wait()  # 等待api.py启动完成
 
-            if p:= processes.get("webui"):
+            if p := processes.get("webui"):
                 p.start()
                 p.name = f"{p.name} ({p.pid})"
                 webui_started.wait()  # 等待webui.py启动完成
@@ -831,7 +835,6 @@ async def start_main_server():
                         else:
                             logger.error(f"未找到模型进程：{model_name}")
 
-
             # for process in processes.get("model_worker", {}).values():
             #     process.join()
             # for process in processes.get("online_api", {}).values():
@@ -866,10 +869,9 @@ async def start_main_server():
             for p in processes.values():
                 logger.info("Process status: %s", p)
 
-if __name__ == "__main__":
-    # 确保数据库表被创建
-    create_tables()
 
+if __name__ == "__main__":
+    create_tables()
     if sys.version_info < (3, 10):
         loop = asyncio.get_event_loop()
     else:
@@ -879,16 +881,15 @@ if __name__ == "__main__":
             loop = asyncio.new_event_loop()
 
         asyncio.set_event_loop(loop)
-    # 同步调用协程代码
-    loop.run_until_complete(start_main_server())
 
+    loop.run_until_complete(start_main_server())
 
 # 服务启动后接口调用示例：
 # import openai
 # openai.api_key = "EMPTY" # Not support yet
 # openai.api_base = "http://localhost:8888/v1"
 
-# model = "chatglm2-6b"
+# model = "chatglm3-6b"
 
 # # create a chat completion
 # completion = openai.ChatCompletion.create(