We don't publish
your competitive advantage.

IFAssertionError 'generation is not None' when streaming with AzureChatOpenAI if the model does not support streaming or the Azure content filter triggers.

THENExplicitly set streaming=False when instantiating AzureChatOpenAI if the deployment does not support streaming. For content filter issues, wrap streaming calls in a try-except block to catch BadRequestError or handle gracefully. Additionally, consider increasing recursion limit or limiting input length as workarounds.

IFUsing stream_options={'include_usage': True} with Groq models causes IndexError in databricks streaming chunk parser due to missing choices in final usage chunk.

THENTo avoid the error, disable include_usage by setting `stream_options: {}` in the model's litellm_params configuration, or omit stream_options from the completion request. This prevents the streaming handler from encountering an empty choices list.

IFUsing litellm.acompletion with stream=True and stream_options={"include_usage": True} on Groq models causes an IndexError: list index out of range in the Databricks streaming parser.

THENRemove the stream_options parameter entirely or set it to an empty dictionary in the model configuration to avoid the crash. Alternatively, disable include_usage.

IFAssertionError on 'assert generation is not None' when calling .stream() on an AzureChatOpenAI model that does not support streaming.

THENCheck if your AzureChatOpenAI deployment supports streaming. If not, avoid using .stream() and instead use .invoke() or .batch(). Alternatively, instantiate the model with streaming=False if the class accepts that parameter. For LangChain core, consider implementing a fallback to non-streaming when the model lacks streaming capability.