scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-09-11 22:51:28 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 123

Fix pydantic serialization warning for ResponseAPIUsage in streaming logging

Browse source 
In _get_assembled_streaming_response, usage was transformed from
ResponseAPIUsage to Chat Completion format and set as a raw dict via
setattr. This bypassed pydantic validation, so subsequent model_dump()
calls on ResponsesAPIResponse emitted a PydanticSerializationUnexpectedValue
warning on every streaming LLM call.

Fix: wrap the transformed usage in a proper ResponseAPIUsage instance
(which allows extra fields) instead of setting a raw dict. The Chat
Completion keys (prompt_tokens, completion_tokens, etc.) are preserved
as extra fields for downstream logging consumers.
This commit is contained in:
Guilherme Pires 2026-02-13 12:50:47 -08:00
parent a06113ec82
commit bb18337562
1 changed files with 17 additions and 5 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

22
litellm/litellm_core_utils/litellm_logging.py
View file

@ -3226,14 +3226,26 @@ class Logging(LiteLLMLoggingBaseClass):
result.response.usage
)
)
# Set as dict instead of Usage object so model_dump() serializes it correctly
# Transform usage to Chat Completion format for internal logging,
# but wrap it in a ResponseAPIUsage so that model_dump() on
# ResponsesAPIResponse serializes cleanly without pydantic warnings.
# ResponseAPIUsage has model_config = {"extra": "allow"}, so Chat
# Completion keys (prompt_tokens, completion_tokens, etc.) are
# preserved as extra fields for downstream logging consumers.
usage_dict = (
transformed_usage.model_dump()
if hasattr(transformed_usage, "model_dump")
else dict(transformed_usage)
)
setattr(
result.response,
"usage",
(
transformed_usage.model_dump()
if hasattr(transformed_usage, "model_dump")
else dict(transformed_usage)
ResponseAPIUsage(
input_tokens=usage_dict.get("prompt_tokens", 0) or 0,
output_tokens=usage_dict.get("completion_tokens", 0) or 0,
total_tokens=usage_dict.get("total_tokens", 0) or 0,
**{k: v for k, v in usage_dict.items()
if k not in ("prompt_tokens", "completion_tokens", "total_tokens")},
),
)
return result.response