v0.2.1

Fixed

• Streaming no longer crashes on chunks with empty choices (e.g. the usage-only final chunk emitted when stream_options={"include_usage": True}). Thanks @MrDiggles2! (#27)
• stream_options is no longer sent on non-streaming requests. The inference API rejects it unless stream=True, so invoke() on models configured with stream_options failed with a 400 validation error. usage_metadata is still populated from the response. (#29)
• user_agent_version now reports the actual installed package version instead of a hardcoded stale one. (#30)

Changed

• Updated integration tests for the current inference API: replaced retired models with currently served ones; empty prompts are now accepted; raised the o3-mini test token budget. (#28)

Install: pip install langchain-gradient==0.2.1