diff --git a/pkg/provider/modelmgr/requesters/chatcmpl.py b/pkg/provider/modelmgr/requesters/chatcmpl.py
index adeaa251..2d2a0b7e 100644
--- a/pkg/provider/modelmgr/requesters/chatcmpl.py
+++ b/pkg/provider/modelmgr/requesters/chatcmpl.py
@@ -160,7 +160,7 @@ class OpenAIChatCompletions(requester.ProviderAPIRequester):
thinking_started = False
thinking_ended = False
role = 'assistant' # 默认角色
- accumulated_reasoning = '' # 仅用于判断何时结束思维链
+ # accumulated_reasoning = '' # 仅用于判断何时结束思维链
async for chunk in self._req_stream(args, extra_body=extra_args):
# 解析 chunk 数据
@@ -182,7 +182,7 @@ class OpenAIChatCompletions(requester.ProviderAPIRequester):
# 处理 reasoning_content
if reasoning_content:
- accumulated_reasoning += reasoning_content
+ # accumulated_reasoning += reasoning_content
# 如果设置了 remove_think，跳过 reasoning_content
if remove_think:
chunk_idx += 1
@@ -289,6 +289,7 @@ class OpenAIChatCompletions(requester.ProviderAPIRequester):
# 发送请求
resp = await self._req(args, extra_body=extra_args)
+ print(resp)
# 处理请求结果
message = await self._make_msg(resp, remove_think)
diff --git a/pkg/provider/modelmgr/requesters/giteeaichatcmpl.py b/pkg/provider/modelmgr/requesters/giteeaichatcmpl.py
index 0ff49798..f8cf15ca 100644
--- a/pkg/provider/modelmgr/requesters/giteeaichatcmpl.py
+++ b/pkg/provider/modelmgr/requesters/giteeaichatcmpl.py
@@ -3,7 +3,7 @@ from __future__ import annotations
import typing
-from . import chatcmpl
+from . import ppiochatcmpl
from .. import requester
from ....core import entities as core_entities
from ... import entities as llm_entities
@@ -12,7 +12,7 @@ import re
import openai.types.chat.chat_completion as chat_completion
-class GiteeAIChatCompletions(chatcmpl.OpenAIChatCompletions):
+class GiteeAIChatCompletions(ppiochatcmpl.PPIOChatCompletions):
"""Gitee AI ChatCompletions API 请求器"""
default_config: dict[str, typing.Any] = {
@@ -20,181 +20,3 @@ class GiteeAIChatCompletions(chatcmpl.OpenAIChatCompletions):
'timeout': 120,
}
- async def _closure(
- self,
- query: core_entities.Query,
- req_messages: list[dict],
- use_model: requester.RuntimeLLMModel,
- use_funcs: list[tools_entities.LLMFunction] = None,
- extra_args: dict[str, typing.Any] = {},
- remove_think: bool = False,
- ) -> llm_entities.Message:
- self.client.api_key = use_model.token_mgr.get_token()
-
- args = {}
- args['model'] = use_model.model_entity.name
-
- if use_funcs:
- tools = await self.ap.tool_mgr.generate_tools_for_openai(use_funcs)
-
- if tools:
- args['tools'] = tools
-
- # gitee 不支持多模态，把content都转换成纯文字
- for m in req_messages:
- if 'content' in m and isinstance(m['content'], list):
- m['content'] = ' '.join([c['text'] for c in m['content']])
-
- args['messages'] = req_messages
-
- resp = await self._req(args, extra_body=extra_args)
-
-
- message = await self._make_msg(resp, remove_think)
-
- return message
-
- async def _make_msg(
- self,
- chat_completion: chat_completion.ChatCompletion,
- remove_think: bool,
- ) -> llm_entities.Message:
- chatcmpl_message = chat_completion.choices[0].message.model_dump()
- # print(chatcmpl_message.keys(), chatcmpl_message.values())
-
- # 确保 role 字段存在且不为 None
- if 'role' not in chatcmpl_message or chatcmpl_message['role'] is None:
- chatcmpl_message['role'] = 'assistant'
-
- reasoning_content = chatcmpl_message['reasoning_content'] if 'reasoning_content' in chatcmpl_message else None
-
- # deepseek的reasoner模型
- if remove_think:
- chatcmpl_message['content'] = re.sub(
- r'.*?', '', chatcmpl_message['content'], flags=re.DOTALL
- )
- else:
- if reasoning_content is not None:
- chatcmpl_message['content'] = (
- '\n' + reasoning_content + '\n\n' + chatcmpl_message['content']
- )
-
- message = llm_entities.Message(**chatcmpl_message)
-
- return message
-
- async def _make_msg_chunk(
- self,
- delta: dict[str, typing.Any],
- idx: int,
- ) -> llm_entities.MessageChunk:
- # 处理流式chunk和完整响应的差异
- # print(chat_completion.choices[0])
-
-
- # 确保 role 字段存在且不为 None
- if 'role' not in delta or delta['role'] is None:
- delta['role'] = 'assistant'
-
- reasoning_content = delta['reasoning_content'] if 'reasoning_content' in delta else None
-
- delta['content'] = '' if delta['content'] is None else delta['content']
- # print(reasoning_content)
-
- # deepseek的reasoner模型
-
- if reasoning_content is not None:
- delta['content'] += reasoning_content
-
- message = llm_entities.MessageChunk(**delta)
-
- return message
-
- async def _closure_stream(
- self,
- query: core_entities.Query,
- req_messages: list[dict],
- use_model: requester.RuntimeLLMModel,
- use_funcs: list[tools_entities.LLMFunction] = None,
- extra_args: dict[str, typing.Any] = {},
- remove_think: bool = False,
- ) -> llm_entities.Message | typing.AsyncGenerator[llm_entities.MessageChunk, None]:
- self.client.api_key = use_model.token_mgr.get_token()
-
- args = {}
- args['model'] = use_model.model_entity.name
-
- if use_funcs:
- tools = await self.ap.tool_mgr.generate_tools_for_openai(use_funcs)
-
- if tools:
- args['tools'] = tools
-
- # 设置此次请求中的messages
- messages = req_messages.copy()
-
- # 检查vision
- for msg in messages:
- if 'content' in msg and isinstance(msg['content'], list):
- for me in msg['content']:
- if me['type'] == 'image_base64':
- me['image_url'] = {'url': me['image_base64']}
- me['type'] = 'image_url'
- del me['image_base64']
-
- args['messages'] = messages
-
- current_content = ''
- args['stream'] = True
- chunk_idx = 0
- is_think = False
- tool_calls_map: dict[str, llm_entities.ToolCall] = {}
- async for chunk in self._req_stream(args, extra_body=extra_args):
- # 处理流式消息
- if hasattr(chunk, 'choices'):
- # 完整响应模式
- if chunk.choices:
- choice = chunk.choices[0]
- delta = choice.delta.model_dump() if hasattr(choice, 'delta') else choice.message.model_dump()
- else:
- continue
- else:
- # 流式chunk模式
- delta = chunk.delta.model_dump() if hasattr(chunk, 'delta') else {}
- if remove_think:
- print(delta)
- if delta['content'] == '':
- is_think = True
- continue
- elif delta['content'] == r'':
- is_think = False
- continue
- elif is_think or delta['content'] == '\n\n':
- continue
-
- delta_message = await self._make_msg_chunk(delta, chunk_idx)
- if delta_message.content:
- current_content += delta_message.content
- delta_message.content = current_content
- # delta_message.all_content = current_content
- if delta_message.tool_calls:
- for tool_call in delta_message.tool_calls:
- if tool_call.id not in tool_calls_map:
- tool_calls_map[tool_call.id] = llm_entities.ToolCall(
- id=tool_call.id,
- type=tool_call.type,
- function=llm_entities.FunctionCall(
- name=tool_call.function.name if tool_call.function else '', arguments=''
- ),
- )
- if tool_call.function and tool_call.function.arguments:
- # 流式处理中，工具调用参数可能分多个chunk返回，需要追加而不是覆盖
- tool_calls_map[tool_call.id].function.arguments += tool_call.function.arguments
-
- chunk_idx += 1
- chunk_choices = getattr(chunk, 'choices', None)
- if chunk_choices and getattr(chunk_choices[0], 'finish_reason', None):
- delta_message.is_final = True
- delta_message.content = current_content
-
- yield delta_message
diff --git a/pkg/provider/modelmgr/requesters/modelscopechatcmpl.py b/pkg/provider/modelmgr/requesters/modelscopechatcmpl.py
index 0007623e..c526313a 100644
--- a/pkg/provider/modelmgr/requesters/modelscopechatcmpl.py
+++ b/pkg/provider/modelmgr/requesters/modelscopechatcmpl.py
@@ -36,6 +36,7 @@ class ModelScopeChatCompletions(requester.ProviderAPIRequester):
self,
args: dict,
extra_body: dict = {},
+ remove_think:bool = False,
) -> chat_completion.ChatCompletion:
args['stream'] = True
@@ -47,11 +48,35 @@ class ModelScopeChatCompletions(requester.ProviderAPIRequester):
resp_gen: openai.AsyncStream = await self.client.chat.completions.create(**args, extra_body=extra_body)
+ chunk_idx = 0
+ thinking_started = False
+ thinking_ended = False
async for chunk in resp_gen:
# print(chunk)
if not chunk or not chunk.id or not chunk.choices or not chunk.choices[0] or not chunk.choices[0].delta:
continue
+ reasoning_content = chunk.choices[0].delta.reasoning_content
+ # 处理 reasoning_content
+ if reasoning_content:
+ # accumulated_reasoning += reasoning_content
+ # 如果设置了 remove_think，跳过 reasoning_content
+ if remove_think:
+ chunk_idx += 1
+ continue
+
+ # 第一次出现 reasoning_content，添加 开始标签
+ if not thinking_started:
+ thinking_started = True
+ pending_content += '\n' + reasoning_content
+ else:
+ # 继续输出 reasoning_content
+ pending_content += reasoning_content
+ elif thinking_started and not thinking_ended and chunk.choices[0].delta.content:
+ # reasoning_content 结束，normal content 开始，添加 结束标签
+ thinking_ended = True
+ pending_content += '\n\n' + chunk.choices[0].delta.content
+
if chunk.choices[0].delta.content is not None:
pending_content += chunk.choices[0].delta.content
@@ -130,6 +155,7 @@ class ModelScopeChatCompletions(requester.ProviderAPIRequester):
use_model: requester.RuntimeLLMModel,
use_funcs: list[tools_entities.LLMFunction] = None,
extra_args: dict[str, typing.Any] = {},
+ remove_think:bool = False,
) -> llm_entities.Message:
self.client.api_key = use_model.token_mgr.get_token()
@@ -157,7 +183,7 @@ class ModelScopeChatCompletions(requester.ProviderAPIRequester):
args['messages'] = messages
# 发送请求
- resp = await self._req(args, extra_body=extra_args)
+ resp = await self._req(args, extra_body=extra_args, remove_think=remove_think)
# 处理请求结果
message = await self._make_msg(resp)
@@ -172,41 +198,6 @@ class ModelScopeChatCompletions(requester.ProviderAPIRequester):
async for chunk in await self.client.chat.completions.create(**args, extra_body=extra_body):
yield chunk
- async def _make_msg_chunk(self,
- delta: dict[str, typing.Any],
- idx: int,
- is_content: bool,
- is_think: bool,
- ) -> llm_entities.MessageChunk:
- # 处理流式chunk和完整响应的差异
- # print(chat_completion.choices[0])
-
- if 'role' not in delta or delta['role'] is None:
- delta['role'] = 'assistant'
-
- reasoning_content = delta['reasoning_content']
-
- delta['content'] = '' if delta['content'] is None else delta['content']
- # print(reasoning_content)
-
- # deepseek的reasoner模型
-
- if reasoning_content is not None and idx == 0:
- delta['content'] += f'\n{reasoning_content}'
- is_think = True
- elif reasoning_content is None and idx != 0:
- if is_content:
- delta['content'] = delta['content']
- elif is_think:
- delta['content'] = f'\n\n\n{delta["content"]}'
- is_content = True
- is_think = False
- elif reasoning_content is not None:
- delta['content'] = reasoning_content
-
- message = llm_entities.MessageChunk(**delta)
-
- return message, is_content, is_think
async def _closure_stream(
self,
@@ -250,7 +241,7 @@ class ModelScopeChatCompletions(requester.ProviderAPIRequester):
thinking_started = False
thinking_ended = False
role = 'assistant' # 默认角色
- accumulated_reasoning = '' # 仅用于判断何时结束思维链
+ # accumulated_reasoning = '' # 仅用于判断何时结束思维链
async for chunk in self._req_stream(args, extra_body=extra_args):
# 解析 chunk 数据
@@ -272,7 +263,7 @@ class ModelScopeChatCompletions(requester.ProviderAPIRequester):
# 处理 reasoning_content
if reasoning_content:
- accumulated_reasoning += reasoning_content
+ # accumulated_reasoning += reasoning_content
# 如果设置了 remove_think，跳过 reasoning_content
if remove_think:
chunk_idx += 1
@@ -365,7 +356,7 @@ class ModelScopeChatCompletions(requester.ProviderAPIRequester):
try:
return await self._closure(
- query=query, req_messages=req_messages, use_model=model, use_funcs=funcs, extra_args=extra_args
+ query=query, req_messages=req_messages, use_model=model, use_funcs=funcs, extra_args=extra_args, remove_think=remove_think
)
except asyncio.TimeoutError:
raise errors.RequesterError('请求超时')
diff --git a/pkg/provider/modelmgr/requesters/ppiochatcmpl.py b/pkg/provider/modelmgr/requesters/ppiochatcmpl.py
index 49f03143..967bb676 100644
--- a/pkg/provider/modelmgr/requesters/ppiochatcmpl.py
+++ b/pkg/provider/modelmgr/requesters/ppiochatcmpl.py
@@ -39,20 +39,45 @@ class PPIOChatCompletions(chatcmpl.OpenAIChatCompletions):
reasoning_content = chatcmpl_message['reasoning_content'] if 'reasoning_content' in chatcmpl_message else None
# deepseek的reasoner模型
- if remove_think:
- chatcmpl_message['content'] = re.sub(
- r'.*?', '', chatcmpl_message['content'], flags=re.DOTALL
- )
- else:
- if reasoning_content is not None:
- chatcmpl_message['content'] = (
- '\n' + reasoning_content + '\n\n' + chatcmpl_message['content']
- )
+ chatcmpl_message["content"] = await self._process_thinking_content(
+ chatcmpl_message['content'],reasoning_content,remove_think)
+
+ # 移除 reasoning_content 字段，避免传递给 Message
+ if 'reasoning_content' in chatcmpl_message:
+ del chatcmpl_message['reasoning_content']
+
message = llm_entities.Message(**chatcmpl_message)
return message
+ async def _process_thinking_content(
+ self,
+ content: str,
+ reasoning_content: str = None,
+ remove_think: bool = False,
+ ) -> tuple[str, str]:
+ """处理思维链内容
+
+ Args:
+ content: 原始内容
+ reasoning_content: reasoning_content 字段内容
+ remove_think: 是否移除思维链
+
+ Returns:
+ 处理后的内容
+ """
+ if remove_think:
+ content = re.sub(
+ r'.*?', '', content, flags=re.DOTALL
+ )
+ else:
+ if reasoning_content is not None:
+ content = (
+ '\n' + reasoning_content + '\n\n' + content
+ )
+ return content
+
async def _make_msg_chunk(
self,
delta: dict[str, typing.Any],
@@ -119,7 +144,6 @@ class PPIOChatCompletions(chatcmpl.OpenAIChatCompletions):
thinking_started = False
thinking_ended = False
role = 'assistant' # 默认角色
- accumulated_reasoning = '' # 仅用于判断何时结束思维链
async for chunk in self._req_stream(args, extra_body=extra_args):
# 解析 chunk 数据
if hasattr(chunk, 'choices') and chunk.choices:
@@ -140,14 +164,18 @@ class PPIOChatCompletions(chatcmpl.OpenAIChatCompletions):
if remove_think:
if delta['content'] is not None:
- if '' in delta['content']:
- is_think = True
+ if '' in delta['content'] and not thinking_started and not thinking_ended:
+ thinking_started = True
continue
- elif delta['content'] == r'':
- is_think = False
+ elif delta['content'] == r'' and not thinking_ended:
+ thinking_ended = True
continue
- elif is_think or delta['content'] == '\n\n':
+ elif thinking_ended and delta['content'] == '\n\n' and thinking_started:
+ thinking_started = False
continue
+ elif thinking_started and not thinking_ended:
+ continue
+
delta_tool_calls = None
if delta.get('tool_calls'):