1

1

backend/api/schemas/conversation_schemas.py

@@ -26,7 +26,7 @@ def _validate_model(_source: ChatSourceTypes, model: str | None):
         logger.warning(f"model {model} not in openai_api models: {'|'.join(list(OpenaiApiChatModels))}")
 
 
-MAX_CONTEXT_MESSAGE_COUNT = 100
+MAX_CONTEXT_MESSAGE_COUNT = 1000
 
 
 class AskRequest(BaseModel):

backend/api/sources/openai_api.py

@@ -2,10 +2,8 @@
 import uuid
 from datetime import datetime, timezone
 from typing import Optional
-
 import httpx
 from pydantic import ValidationError
-
 from api.conf import Config, Credentials
 from api.enums import OpenaiApiChatModels, ChatSourceTypes
 from api.exceptions import OpenaiApiException
@@ -16,12 +14,9 @@
 from utils.logger import get_logger
 
 logger = get_logger(__name__)
-
 config = Config()
 credentials = Credentials()
-
-MAX_CONTEXT_MESSAGE_COUNT = 100
-
+MAX_CONTEXT_MESSAGE_COUNT = 1000
 
 async def _check_response(response: httpx.Response) -> None:
     # 改成自带的错误处理
@@ -35,7 +30,6 @@ async def _check_response(response: httpx.Response) -> None:
         )
         raise error from ex
 
-
 def make_session() -> httpx.AsyncClient:
     if config.openai_api.proxy is not None:
         proxies = {
@@ -47,12 +41,10 @@ def make_session() -> httpx.AsyncClient:
         session = httpx.AsyncClient(timeout=None)
     return session
 
-
 class OpenaiApiChatManager(metaclass=SingletonMeta):
     """
     OpenAI API Manager
     """
-
     def __init__(self):
         self.session = make_session()
 
@@ -62,9 +54,7 @@ def reset_session(self):
     async def complete(self, model: OpenaiApiChatModels, text_content: str, conversation_id: uuid.UUID = None,
                        parent_message_id: uuid.UUID = None,
                        context_message_count: int = -1, extra_args: Optional[dict] = None, **_kwargs):
-
         assert config.openai_api.enabled, "openai_api is not enabled"
-
         now_time = datetime.now().astimezone(tz=timezone.utc)
         message_id = uuid.uuid4()
         new_message = OpenaiApiChatMessage(
@@ -79,9 +69,7 @@ async def complete(self, model: OpenaiApiChatModels, text_content: str, conversa
                 source="openai_api",
             )
         )
-
         messages = []
-
         if not conversation_id:
             assert parent_message_id is None, "parent_id must be None when conversation_id is None"
             messages = [new_message]
@@ -93,45 +81,29 @@ async def complete(self, model: OpenaiApiChatModels, text_content: str, conversa
                 raise ValueError(f"{conversation_id} is not api conversation")
             if not conv_history.mapping.get(str(parent_message_id)):
                 raise ValueError(f"{parent_message_id} is not a valid parent of {conversation_id}")
-
-            # 从 current_node 开始往前找 context_message_count 个 message
+            # 从 current_node 开始往前找最多 5 个 message
             if not conv_history.current_node:
                 raise ValueError(f"{conversation_id} current_node is None")
-
             msg = conv_history.mapping.get(str(conv_history.current_node))
             assert msg, f"{conv_history.id} current_node({conv_history.current_node}) not found in mapping"
-
             count = 0
-            iter_count = 0
-
-            while msg:
-                count += 1
+            while msg and count < 5:
                 messages.append(msg)
-                if context_message_count != -1 and count >= context_message_count:
-                    break
-                iter_count += 1
-                if iter_count > MAX_CONTEXT_MESSAGE_COUNT:
-                    raise ValueError(f"too many messages to iterate, conversation_id={conversation_id}")
                 msg = conv_history.mapping.get(str(msg.parent))
-
+                count += 1
             messages.reverse()
             messages.append(new_message)
-
         # TODO: credits 判断
-
         base_url = config.openai_api.openai_base_url
         data = {
             "model": model.code(),
             "messages": [{"role": msg.role, "content": msg.content.text} for msg in messages],
             "stream": True,
             **(extra_args or {})
         }
-
         reply_message = None
         text_content = ""
-
         timeout = httpx.Timeout(config.openai_api.read_timeout, connect=config.openai_api.connect_timeout)
-
         async with self.session.stream(method="POST",
                                        url=f"{base_url}chat/completions",
                                        json=data,
@@ -146,14 +118,11 @@ async def complete(self, model: OpenaiApiChatModels, text_content: str, conversa
                     line = line[6:]
                 if "[DONE]" in line:
                     break
-
                 try:
                     line = json.loads(line)
                     resp = OpenaiChatResponse.model_validate(line)
-
                     if not resp.choices or len(resp.choices) == 0:
                         continue
-
                     if resp.choices[0].message is not None:
                         text_content = resp.choices[0].message.get("content")
                     if resp.choices[0].delta is not None:
@@ -175,13 +144,10 @@ async def complete(self, model: OpenaiApiChatModels, text_content: str, conversa
                         )
                     else:
                         reply_message.content = OpenaiApiChatMessageTextContent(content_type="text", text=text_content)
-
                     if resp.usage:
                         reply_message.metadata.usage = resp.usage
-
                     yield reply_message
-
                 except json.decoder.JSONDecodeError:
                     logger.warning(f"OpenAIChatResponse parse json error")
                 except ValidationError as e:
-                    logger.warning(f"OpenAIChatResponse validate error: {e}")
+                    logger.warning(f"OpenAIChatResponse validate error: {e}")