Release langchain-openai==0.1.9
Changes since langchain-openai==0.1.8
get_num_tokens_from_messages
will now estimate token consumption for images as described in OpenAI's documentation here.- Token usage information can now be toggled in streaming mode using the
stream_usage
parameter. - Invoke and streaming responses now include model version metadata. System fingerprint was also added to streaming.
- Option to disable parallel tool calls (
parallel_tool_calls
) documented in API reference. - Misc improvements and bug fixes.
Commits:
openai: release 0.1.9 (#23263)
partners[minor]: Fix value error message for with_structured_output (#22877)
infra: add more formatter rules to openai (#23189)
openai[patch]: image token counting (#23147)
openai[patch], standard-tests[patch]: don't pass in falsey stop vals (#23153)
standard-tests[patch]: Update chat model standard tests (#22378)
openai[patch]: add stream_usage parameter (#22854)
[Partner]: Add metadata to stream response (#22716)
partners: fix numpy dep (#22858)
openai: add parallel_tool_calls
to api ref (#22746)
openai[patch]: correct grammar in exception message in embeddings/base.py (#22629)
openai, azure: update model_name in ChatResult to use name from API response (#22569)
docs: update anthropic chat model (#22483)
openai: update ChatOpenAI api ref (#22324)