You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Describe the feature or improvement you're requesting
Hi there! Could we make the response deserialization more flexible to preserve custom fields like reasoning_content?
When using DeepSeek's API (which is OpenAI-compatible), their R1 model returns a reasoning_content field at:
response.choices[0].message.reasoning_content
But currently these custom fields get silently dropped. Maybe we could keep them like OpenAI's Python SDK does?
Additional context
I checked OpenAI's official Python SDK. It seems that they just deserialised the generated json and didn't perform any checks. So the unknown fields were just preserved by dynamic typing.
Describe the feature or improvement you're requesting
Hi there! Could we make the response deserialization more flexible to preserve custom fields like
reasoning_content
?When using DeepSeek's API (which is OpenAI-compatible), their R1 model returns a
reasoning_content
field at:But currently these custom fields get silently dropped. Maybe we could keep them like OpenAI's Python SDK does?
Additional context
I checked OpenAI's official Python SDK. It seems that they just deserialised the generated json and didn't perform any checks. So the unknown fields were just preserved by dynamic typing.
deepseek's reasoning model docs
The text was updated successfully, but these errors were encountered: