scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-10-09 03:18:44 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 109

[Bug] Fix missing image_tokens in Responses API output_tokens_details (#20404)

Browse source 
When transforming chat completion responses to Responses API format,
image_tokens from completion_tokens_details was not being included
in output_tokens_details. This affected Vertex AI/Gemini models that
return image token counts in candidatesTokensDetails with modality="IMAGE".

The fix adds image_tokens handling alongside existing reasoning_tokens
and text_tokens transformation.
This commit is contained in:
Lei Nie 2026-02-04 23:38:50 -08:00 • committed by Sameer Kankute
parent 1ee43b11de
commit 08824f9912
2 changed files with 47 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
litellm/responses/litellm_completion_transformation/transformation.py
View file

@ -1794,6 +1794,12 @@ class LiteLLMCompletionResponsesConfig:
):
output_details_dict["text_tokens"] = completion_details.text_tokens
if (
hasattr(completion_details, "image_tokens")
and completion_details.image_tokens is not None
):
output_details_dict["image_tokens"] = completion_details.image_tokens
if output_details_dict:
response_usage.output_tokens_details = OutputTokensDetails(
**output_details_dict

41
tests/test_litellm/responses/litellm_completion_transformation/test_litellm_completion_responses.py
View file

@ -1424,6 +1424,47 @@ class TestUsageTransformation:
assert response_usage.input_tokens_details is None
assert response_usage.output_tokens_details is None
def test_transform_usage_with_image_tokens(self):
"""Test that image_tokens from Vertex AI/Gemini are properly transformed to output_tokens_details"""
# Setup: Simulate Vertex AI/Gemini usage with image_tokens in completion_tokens_details
usage = Usage(
prompt_tokens=10,
completion_tokens=150,
total_tokens=160,
completion_tokens_details=CompletionTokensDetailsWrapper(
reasoning_tokens=0,
text_tokens=50,
image_tokens=100, # From Vertex AI candidatesTokensDetails with modality="IMAGE"
),
)
chat_completion_response = ModelResponse(
id="test-response-id",
created=1234567890,
model="gemini-2.0-flash",
object="chat.completion",
usage=usage,
choices=[
Choices(
finish_reason="stop",
index=0,
message=Message(content="Here is the generated image.", role="assistant"),
)
],
)
# Execute
response_usage = LiteLLMCompletionResponsesConfig._transform_chat_completion_usage_to_responses_usage(
chat_completion_response=chat_completion_response
)
# Assert
assert response_usage.output_tokens == 150
assert response_usage.output_tokens_details is not None
assert response_usage.output_tokens_details.reasoning_tokens == 0
assert response_usage.output_tokens_details.text_tokens == 50
assert response_usage.output_tokens_details.image_tokens == 100
class TestStreamingIDConsistency:
"""Test cases for consistent IDs across streaming events (issue #14962)"""