scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-09-19 00:01:29 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 88

fix(cost): avoid charging audio duration and tokens twice

Browse source
This commit is contained in:
moe-berri 2026-09-14 18:04:31 -07:00
parent 8b53d48e34
commit f79a413e74
2 changed files with 24 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
litellm/litellm_core_utils/llm_cost_calc/utils.py
View file

@ -956,7 +956,9 @@ def _calculate_input_cost(
)
### AUDIO COST
if prompt_tokens_details["audio_tokens"]:
if prompt_tokens_details["audio_tokens"] and not (
prompt_tokens_details["audio_length_seconds"] and model_info.get("input_cost_per_audio_per_second") is not None
):
audio_cost_key: Final = _get_service_tier_cost_key("input_cost_per_audio_token", service_tier)
prompt_cost += calculate_cost_component(model_info, audio_cost_key, prompt_tokens_details["audio_tokens"])

21
tests/test_litellm/litellm_core_utils/llm_cost_calc/test_llm_cost_calc_utils.py
View file

@ -1,5 +1,6 @@
import json
from datetime import datetime, timezone
from typing import Final
import pytest
from fastapi.testclient import TestClient
@ -52,6 +53,26 @@ def _local_model_cost_map(monkeypatch: pytest.MonkeyPatch) -> None:
monkeypatch.setattr(litellm, "model_cost", litellm.get_model_cost_map(url=""))
@pytest.mark.parametrize(
"duration,second_rate,expected",
((2.0, 0.00016, 0.00032), (2.0, 0.0, 0.0), (2.0, None, 0.000416), (0.0, 0.00016, 0.000416)),
)
def test_audio_duration_and_tokens_bill_only_once(duration: float, second_rate: float | None, expected: float) -> None:
info: Final[ModelInfo] = {
"input_cost_per_token": 0.0,
"output_cost_per_token": 0.0,
"input_cost_per_audio_token": 6.5e-6,
"input_cost_per_audio_per_second": second_rate,
}
usage: Final = Usage(
prompt_tokens=64,
completion_tokens=0,
prompt_tokens_details=PromptTokensDetailsWrapper(audio_tokens=64, audio_length_seconds=duration, text_tokens=0),
)
cost, _ = generic_cost_per_token("audio-billing-fixture", usage, "vertex_ai", model_info=info)
assert cost == pytest.approx(expected)
@pytest.mark.parametrize("prompt_tokens", [100, 200000, 200001])
@pytest.mark.parametrize("read_rate", [None, 0.0, 0.25e-6])
@pytest.mark.parametrize("service_tier", [None, "priority"])