mirror of
https://github.com/hwchase17/langchain.git
synced 2026-06-09 18:50:33 +00:00
50fa524a6d
## Description
deepseek-r1 always returns an empty string `reasoning_content` to the
first chunk when thinking, and sets `reasoning_content` to None when
thinking is over, to determine when to switch to normal output.
Therefore, whether the reasoning_content field exists should be judged
as None.
## Demo
deepseek-r1 reasoning output:
```
{'delta': {'content': None, 'function_call': None, 'refusal': None, 'role': 'assistant', 'tool_calls': None, 'reasoning_content': ''}, 'finish_reason': None, 'index': 0, 'logprobs': None}
{'delta': {'content': None, 'function_call': None, 'refusal': None, 'role': None, 'tool_calls': None, 'reasoning_content': '好的'}, 'finish_reason': None, 'index': 0, 'logprobs': None}
{'delta': {'content': None, 'function_call': None, 'refusal': None, 'role': None, 'tool_calls': None, 'reasoning_content': '，'}, 'finish_reason': None, 'index': 0, 'logprobs': None}
{'delta': {'content': None, 'function_call': None, 'refusal': None, 'role': None, 'tool_calls': None, 'reasoning_content': '用户'}, 'finish_reason': None, 'index': 0, 'logprobs': None}
...
```
deepseek-r1 first normal output
```
...
{'delta': {'content': ' main', 'function_call': None, 'refusal': None, 'role': None, 'tool_calls': None, 'reasoning_content': None}, 'finish_reason': None, 'index': 0, 'logprobs': None}
{'delta': {'content': '\n\nimport', 'function_call': None, 'refusal': None, 'role': None, 'tool_calls': None, 'reasoning_content': None}, 'finish_reason': None, 'index': 0, 'logprobs': None}
...
```
---------
Co-authored-by: ccurme <chester.curme@gmail.com>
langchain-deepseek
This package contains the LangChain integration with the DeepSeek API
Installation
pip install -U langchain-deepseek
And you should configure credentials by setting the following environment variables:
DEEPSEEK_API_KEY
Chat Models
ChatDeepSeek class exposes chat models from DeepSeek.
from langchain_deepseek import ChatDeepSeek
llm = ChatDeepSeek(model="deepseek-chat")
llm.invoke("Sing a ballad of LangChain.")