outman/dify
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat:add think tag display for xinference deepseek r1 (#13291)

Browse Source
This commit is contained in:
呆萌闷油瓶 2025-02-06 22:04:58 +08:00 committed by GitHub
parent ba91f34636
commit f7e7a399d9
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 16 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
api/core/model_runtime/model_providers/xinference/llm/llm.py
View File

@ -1,3 +1,4 @@
import re
from collections.abc import Generator, Iterator from collections.abc import Generator, Iterator
from typing import Optional, cast from typing import Optional, cast
@ -635,16 +636,16 @@ class XinferenceAILargeLanguageModel(LargeLanguageModel):
handle stream chat generate response handle stream chat generate response
""" """
full_response = "" full_response = ""
is_reasoning_started_tag = False
for chunk in resp: for chunk in resp:
if len(chunk.choices) == 0: if len(chunk.choices) == 0:
continue continue
delta = chunk.choices[0] delta = chunk.choices[0]
if delta.finish_reason is None and (delta.delta.content is None or delta.delta.content == ""): if delta.finish_reason is None and (delta.delta.content is None or delta.delta.content == ""):
continue continue
delta_content = delta.delta.content
if not delta_content:
delta_content = ""
# check if there is a tool call in the response # check if there is a tool call in the response
function_call = None function_call = None
tool_calls = [] tool_calls = []
@ -657,9 +658,18 @@ class XinferenceAILargeLanguageModel(LargeLanguageModel):
if function_call: if function_call:
assistant_message_tool_calls += [self._extract_response_function_call(function_call)] assistant_message_tool_calls += [self._extract_response_function_call(function_call)]
if not is_reasoning_started_tag and "<think>" in delta_content:
is_reasoning_started_tag = True
delta_content = "> 💭 " + delta_content.replace("<think>", "")
elif is_reasoning_started_tag and "</think>" in delta_content:
delta_content = delta_content.replace("</think>", "") + "\n\n"
is_reasoning_started_tag = False
elif is_reasoning_started_tag:
if "\n" in delta_content:
delta_content = re.sub(r"\n(?!(>|\n))", "\n> ", delta_content)
# transform assistant message to prompt message # transform assistant message to prompt message
assistant_prompt_message = AssistantPromptMessage( assistant_prompt_message = AssistantPromptMessage(
content=delta.delta.content or "", tool_calls=assistant_message_tool_calls content=delta_content or "", tool_calls=assistant_message_tool_calls
) )
if delta.finish_reason is not None: if delta.finish_reason is not None:
@ -697,7 +707,7 @@ class XinferenceAILargeLanguageModel(LargeLanguageModel):
), ),
) )
full_response += delta.delta.content full_response += delta_content
def _handle_completion_generate_response( def _handle_completion_generate_response(
self, self,