feat: add note for remove_think & remove dify remove cot code

This commit is contained in:
Junyan Qin
2025-08-21 21:38:58 +08:00
parent df524b8a7a
commit 87ecb4e519
5 changed files with 17 additions and 66 deletions

View File

@@ -20,7 +20,7 @@ class DBMigratePipelineRemoveCotConfig(migration.DBMigration):
config = serialized_pipeline['config']
if 'remove-think' not in config['output']['misc']:
config['output']['misc']['remove-think'] = True
config['output']['misc']['remove-think'] = False
await self.ap.persistence_mgr.execute_async(
sqlalchemy.update(persistence_pipeline.LegacyPipeline)

View File

@@ -3,7 +3,6 @@ from __future__ import annotations
import typing
import json
import uuid
import re
import base64
@@ -38,33 +37,9 @@ class DifyServiceAPIRunner(runner.RequestRunner):
base_url=self.pipeline_config['ai']['dify-service-api']['base-url'],
)
def _try_convert_thinking(self, resp_text: str) -> str:
"""尝试转换 Dify 的思考提示"""
if not resp_text.startswith(
'<details style="color:gray;background-color: #f8f8f8;padding: 8px;border-radius: 4px;" open> <summary> Thinking... </summary>'
):
return resp_text
if self.pipeline_config['ai']['dify-service-api']['thinking-convert'] == 'original':
return resp_text
if self.pipeline_config['ai']['dify-service-api']['thinking-convert'] == 'remove':
return re.sub(
r'<details style="color:gray;background-color: #f8f8f8;padding: 8px;border-radius: 4px;" open> <summary> Thinking... </summary>.*?</details>',
'',
resp_text,
flags=re.DOTALL,
)
if self.pipeline_config['ai']['dify-service-api']['thinking-convert'] == 'plain':
pattern = r'<details style="color:gray;background-color: #f8f8f8;padding: 8px;border-radius: 4px;" open> <summary> Thinking... </summary>(.*?)</details>'
thinking_text = re.search(pattern, resp_text, flags=re.DOTALL)
content_text = re.sub(pattern, '', resp_text, flags=re.DOTALL)
return f'<think>{thinking_text.group(1)}</think>\n{content_text}'
def _process_thinking_content(
self,
content: str,
self,
content: str,
) -> tuple[str, str]:
"""处理思维链内容
@@ -354,8 +329,9 @@ class DifyServiceAPIRunner(runner.RequestRunner):
yield msg
async def _chat_messages_chunk(self, query: core_entities.Query) -> typing.AsyncGenerator[llm_entities.MessageChunk, None]:
async def _chat_messages_chunk(
self, query: core_entities.Query
) -> typing.AsyncGenerator[llm_entities.MessageChunk, None]:
"""调用聊天助手"""
cov_id = query.session.using_conversation.uuid or ''
query.variables['conversation_id'] = cov_id
@@ -371,8 +347,6 @@ class DifyServiceAPIRunner(runner.RequestRunner):
for image_id in image_ids
]
mode = 'basic' # 标记是基础编排还是工作流编排
basic_mode_pending_chunk = ''
inputs = {}
@@ -411,6 +385,7 @@ class DifyServiceAPIRunner(runner.RequestRunner):
continue
if '</think>' in chunk['answer'] and not think_end:
import re
content = re.sub(r'^\n</think>', '', chunk['answer'])
basic_mode_pending_chunk += content
think_end = True
@@ -433,13 +408,11 @@ class DifyServiceAPIRunner(runner.RequestRunner):
is_final=is_final,
)
if chunk is None:
raise errors.DifyAPIError('Dify API 没有返回任何响应请检查网络连接和API配置')
query.session.using_conversation.uuid = chunk['conversation_id']
async def _agent_chat_messages_chunk(
self, query: core_entities.Query
) -> typing.AsyncGenerator[llm_entities.MessageChunk, None]:
@@ -496,6 +469,7 @@ class DifyServiceAPIRunner(runner.RequestRunner):
continue
if '</think>' in chunk['answer'] and not think_end:
import re
content = re.sub(r'^\n</think>', '', chunk['answer'])
pending_agent_message += content
think_end = True
@@ -509,7 +483,6 @@ class DifyServiceAPIRunner(runner.RequestRunner):
elif chunk['event'] == 'message_end':
is_final = True
else:
if chunk['event'] == 'agent_thought':
if chunk['tool'] != '' and chunk['observation'] != '': # 工具调用结果,跳过
continue
@@ -543,7 +516,6 @@ class DifyServiceAPIRunner(runner.RequestRunner):
role='assistant',
content=[llm_entities.ContentElement.from_image_url(image_url)],
is_final=is_final,
)
if chunk['event'] == 'error':
@@ -560,7 +532,9 @@ class DifyServiceAPIRunner(runner.RequestRunner):
query.session.using_conversation.uuid = chunk['conversation_id']
async def _workflow_messages_chunk(self, query: core_entities.Query) -> typing.AsyncGenerator[llm_entities.MessageChunk, None]:
async def _workflow_messages_chunk(
self, query: core_entities.Query
) -> typing.AsyncGenerator[llm_entities.MessageChunk, None]:
"""调用工作流"""
if not query.session.using_conversation.uuid:
@@ -618,6 +592,7 @@ class DifyServiceAPIRunner(runner.RequestRunner):
continue
if '</think>' in chunk['data']['text'] and not think_end:
import re
content = re.sub(r'^\n</think>', '', chunk['data']['text'])
workflow_contents += content
think_end = True
@@ -650,7 +625,6 @@ class DifyServiceAPIRunner(runner.RequestRunner):
yield msg
if messsage_idx % 8 == 0 or is_final:
yield llm_entities.MessageChunk(
role='assistant',
@@ -694,4 +668,4 @@ class DifyServiceAPIRunner(runner.RequestRunner):
else:
raise errors.DifyAPIError(
f'不支持的 Dify 应用类型: {self.pipeline_config["ai"]["dify-service-api"]["app-type"]}'
)
)

View File

@@ -51,7 +51,6 @@
"base-url": "https://api.dify.ai/v1",
"app-type": "chat",
"api-key": "your-api-key",
"thinking-convert": "plain",
"timeout": 30
},
"dashscope-app-api": {
@@ -88,7 +87,7 @@
"at-sender": true,
"quote-origin": true,
"track-function-calls": false,
"remove-think": true
"remove-think": false
}
}
}

View File

@@ -118,28 +118,6 @@ stages:
zh_Hans: API 密钥
type: string
required: true
- name: thinking-convert
label:
en_US: CoT Convert
zh_Hans: 思维链转换策略
type: select
required: true
default: plain
options:
- name: plain
label:
en_US: Convert to <think>...</think>
zh_Hans: 转换成 <think>...</think>
- name: original
label:
en_US: Original
zh_Hans: 原始
- name: remove
label:
en_US: Remove
zh_Hans: 移除
- name: dashscope-app-api
label:
en_US: Aliyun Dashscope App API

View File

@@ -110,8 +110,8 @@ stages:
en_US: Remove CoT
zh_Hans: 删除思维链
description:
en_US: If enabled, LangBot will remove the LLM thought content in response
zh_Hans: 如果启用,将自动删除大模型回复中的模型思考内容
en_US: 'If enabled, LangBot will remove the LLM thought content in response. Note: When using streaming response, removing CoT may cause the first token to wait for a long time.'
zh_Hans: '如果启用,将自动删除大模型回复中的模型思考内容。注意:当您使用流式响应时,删除思维链可能会导致首个 Token 的等待时间过长'
type: boolean
required: true
default: true
default: false