scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-09-15 23:31:29 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 213

test: regression test for completion_tokens_details on Responses API usage (#15377)

Browse source 
Adds a focused regression test asserting that
ResponseAPILoggingUtils._transform_response_api_usage_to_chat_usage()
populates Usage.completion_tokens_details.reasoning_tokens from
output_tokens_details.reasoning_tokens, matching what Chat Completions
produces.

The underlying fix already shipped on main (shared OpenAI Responses path
used by Azure Responses), so this locks in the behaviour to prevent
regressions on the Azure Responses API path reported in #15377.
This commit is contained in:
mateo-berri 2026-04-23 17:30:56 -07:00
parent 87e120d958
commit 4f0a780391
1 changed files with 32 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

32
tests/test_litellm/responses/test_responses_utils.py
View file

@ -348,6 +348,38 @@ class TestResponseAPILoggingUtils:
assert result.completion_tokens_details.image_tokens == 100
assert result.completion_tokens_details.text_tokens == 70
def test_transform_response_api_usage_populates_reasoning_tokens_regression_15377(
self,
):
"""Regression test for issue #15377.
Azure OpenAI Responses API returns ``usage.output_tokens_details`` with
``reasoning_tokens``. LiteLLM must transform this into
``completion_tokens_details.reasoning_tokens`` on the returned ``Usage``
object so downstream callbacks (e.g. ``async_log_success_event``) see
the reasoning tokens rather than ``completion_tokens_details=None``.
"""
# Raw usage shape as returned by Azure Responses API for a reasoning model.
usage = {
"input_tokens": 12,
"input_tokens_details": {"cached_tokens": 0},
"output_tokens": 204,
"output_tokens_details": {"reasoning_tokens": 64},
"total_tokens": 216,
}
result = ResponseAPILoggingUtils._transform_response_api_usage_to_chat_usage(
usage
)
assert isinstance(result, Usage)
# Core fix: completion_tokens_details must be populated, not None.
assert result.completion_tokens_details is not None
assert result.completion_tokens_details.reasoning_tokens == 64
# Sanity check: prompt_tokens_details still behaves as before.
assert result.prompt_tokens_details is not None
assert result.prompt_tokens_details.cached_tokens == 0
class TestResponsesAPIProviderSpecificParams:
"""