scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-11-02 07:13:02 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 48

fix(cost): limit reported usage costs to non-streaming calls

Browse source
This commit is contained in:
AoHanBei 2026-10-20 22:22:55 +08:00
parent 998a12086a
commit 5dda53612e
2 changed files with 64 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
litellm/cost_calculator.py
View file

@ -2255,6 +2255,8 @@ def response_cost_calculator(
if (
isinstance(response_object, ModelResponse)
and custom_llm_provider == "openai"
and litellm_logging_obj is not None
and litellm_logging_obj.stream is False
and not custom_pricing
and base_model is None
):

62
tests/unit/test_cost_calculator.py
View file

@ -10417,6 +10417,15 @@ def test_openai_reported_usage_cost_respects_pricing_policy(
custom_pricing=pricing_policy == "custom",
base_model=base_model if pricing_policy == "base" else None,
cache_hit=pricing_policy == "cache",
litellm_logging_obj=Logging(
model=model,
messages=[],
stream=False,
call_type="completion",
start_time=datetime.datetime(2026, 1, 1),
litellm_call_id="reported-cost-pricing-policy",
function_id="reported-cost-pricing-policy",
),
)
token_cost: Final = (100 * input_rate + 20 * output_rate) * (2 if pricing_policy == "base" else 1)
expected: Final = (
@ -10448,5 +10457,58 @@ def test_openai_reported_usage_cost_applies_discount_before_margin(
custom_llm_provider="openai",
call_type="completion",
optional_params={},
litellm_logging_obj=Logging(
model="openai/reported-cost-with-margin",
messages=[],
stream=False,
call_type="completion",
start_time=datetime.datetime(2026, 1, 1),
litellm_call_id="reported-cost-margin",
function_id="reported-cost-margin",
),
)
assert cost == pytest.approx(reported_cost * (1 - discount) * (1 + margin) + fixed_amount)
@pytest.mark.parametrize("stream_context", ["absent", "streaming", "unspecified", "non_streaming"])
def test_openai_reported_usage_cost_requires_non_streaming_context(
monkeypatch: pytest.MonkeyPatch, stream_context: str
) -> None:
model: Final = "openai/reported-usage-stream-context"
input_rate: Final = 1e-6
output_rate: Final = 2e-6
reported_cost: Final = 0.037
monkeypatch.setitem(
litellm.model_cost,
model,
{
"litellm_provider": "openai",
"mode": "chat",
"input_cost_per_token": input_rate,
"output_cost_per_token": output_rate,
},
)
logging_obj: Final = (
None
if stream_context == "absent"
else Logging(
model=model,
messages=[],
stream=False if stream_context == "non_streaming" else True if stream_context == "streaming" else None,
call_type="completion",
start_time=datetime.datetime(2026, 1, 1),
litellm_call_id="reported-cost-stream-context",
function_id="reported-cost-stream-context",
)
)
response: Final = ModelResponse(model=model, usage=Usage(prompt_tokens=10, completion_tokens=5, cost=reported_cost))
cost: Final = response_cost_calculator(
response_object=response,
model=model,
custom_llm_provider="openai",
call_type="completion",
optional_params={},
litellm_logging_obj=logging_obj,
)
expected: Final = reported_cost if stream_context == "non_streaming" else 10 * input_rate + 5 * output_rate
assert cost == pytest.approx(expected)