observability-api-response-handlin-langfuse-sdk-logs-pydantic-v1-validation-errors-fo-9041a0fd

THEN Ensure that the usage object returned by OpenAI contains non-empty dicts with integer values for all nested fields (e.g., audio_tokens, accepted_prediction_tokens). If necessary, transform the response to fill missing fields with 0 before passing to langfuse, or update to a newer version of the SDK that handles this case.

# Transform usage to avoid validation errors
if not usage.prompt_tokens_details:
    usage.prompt_tokens_details = {"cached_tokens": 0, "audio_tokens": 0}
if not usage.completion_tokens_details:
    usage.completion_tokens_details = {"reasoning_tokens": 0, "audio_tokens": 0, "accepted_prediction_tokens": 0, "rejected_prediction_tokens": 0}