scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-10-03 02:22:24 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 81

fix: prefer registered video cost limits over metadata

Browse source
This commit is contained in:
pragnyanramtha 2026-05-19 06:40:39 +00:00
parent 0e0c6bf6f3
commit 2cbde0916b
2 changed files with 75 additions and 10 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

23
litellm/cost_calculator.py
View file

@ -316,7 +316,7 @@ def _is_positive_finite_number(value: Any) -> bool:
)
def _get_metadata_model_infos(
def _get_metadata_model_infos_for_cost_fallback(
litellm_logging_obj: Optional[LitellmLoggingObject],
) -> List[Mapping[str, Any]]:
litellm_params = getattr(litellm_logging_obj, "litellm_params", None)
@ -339,15 +339,6 @@ def _get_max_input_tokens_for_cost_fallback(
custom_llm_provider: Optional[str],
litellm_logging_obj: Optional[LitellmLoggingObject],
) -> Optional[Union[int, float]]:
metadata_model_infos = _get_metadata_model_infos(
litellm_logging_obj=litellm_logging_obj
)
for model_info in metadata_model_infos:
for token_limit_key in ("max_input_tokens", "max_tokens"):
token_limit = model_info.get(token_limit_key)
if _is_positive_finite_number(token_limit):
return cast(Union[int, float], token_limit)
if model is None:
return None
@ -369,6 +360,18 @@ def _get_max_input_tokens_for_cost_fallback(
token_limit = model_info.get(token_limit_key)
if _is_positive_finite_number(token_limit):
return cast(Union[int, float], token_limit)
# Request metadata can be caller-controlled on proxy paths. Prefer the
# registered model info above; use metadata only as a final compatibility
# fallback for router/deployment paths that have not registered the model.
metadata_model_infos = _get_metadata_model_infos_for_cost_fallback(
litellm_logging_obj=litellm_logging_obj
)
for model_info in metadata_model_infos:
for token_limit_key in ("max_input_tokens", "max_tokens"):
token_limit = model_info.get(token_limit_key)
if _is_positive_finite_number(token_limit):
return cast(Union[int, float], token_limit)
return None

62
tests/local_testing/test_completion_cost.py
View file

@ -1031,6 +1031,68 @@ def test_completion_cost_uses_conservative_video_fallback_without_usage():
assert cost == pytest.approx(max_input_tokens * input_cost_per_token)
@pytest.mark.parametrize("metadata_key", ["metadata", "litellm_metadata"])
def test_completion_cost_ignores_client_metadata_for_video_fallback_limit(
metadata_key,
):
model = "openai/test-video-untrusted-metadata-fallback"
input_cost_per_token = 0.25
max_input_tokens = 32
litellm.register_model(
model_cost={
model: {
"input_cost_per_token": input_cost_per_token,
"output_cost_per_token": 0.0,
"max_tokens": max_input_tokens,
"max_input_tokens": max_input_tokens,
"max_output_tokens": 4,
"litellm_provider": "openai",
"mode": "chat",
}
}
)
messages = [
{
"role": "user",
"content": [
{
"type": "video_url",
"video_url": {
"url": "https://example.com/video.mp4",
"video_metadata": {
"duration_seconds": 0,
"fps": 0,
"has_audio": False,
},
},
}
],
}
]
logging_obj = MagicMock()
logging_obj.litellm_params = {
metadata_key: {
"model_info": {
"max_input_tokens": 1,
"max_tokens": 1,
}
}
}
try:
cost = completion_cost(
completion_response={"model": model, "usage": {}},
model=model,
messages=messages,
custom_llm_provider="openai",
litellm_logging_obj=logging_obj,
)
finally:
litellm.model_cost.pop(model, None)
assert cost == pytest.approx(max_input_tokens * input_cost_per_token)
def test_completion_cost_uses_provider_video_usage_when_present():
model = "openai/test-video-provider-usage"
input_cost_per_token = 0.25