fix(realtime): bill a translation rate the deployment omits from the session model

This commit is contained in:
mateo-berri 2026-10-28 13:22:32 -07:00
parent 8743521f73
commit f30ea4c7f8
2 changed files with 26 additions and 7 deletions

View file

@ -3758,13 +3758,8 @@ def handle_realtime_translation_cost_calculation(
input_seconds, output_seconds = _translation_session_seconds(results)
if input_seconds <= 0 and output_seconds <= 0:
return 0.0
deployment_rates: Final = _translation_per_second_rates(
(*deployment_pricing_models, litellm_model_name), custom_llm_provider
)
input_rate, output_rate = (
deployment_rates
if any(rate is not None for rate in deployment_rates)
else _translation_per_second_rates((session_model,), custom_llm_provider)
input_rate, output_rate = _translation_per_second_rates(
(*deployment_pricing_models, litellm_model_name, session_model), custom_llm_provider
)
return input_seconds * (input_rate or 0.0) + output_seconds * (output_rate or 0.0)

View file

@ -10332,6 +10332,30 @@ def test_realtime_translation_deployment_per_second_rates_fall_back_to_public_ra
) == pytest.approx(60 * 0.001 + (60 * public_input_rate if bills_public_input_rate else 0.0))
def test_realtime_translation_custom_deployment_with_one_rate_bills_the_session_model_for_the_other(
_local_model_cost_map: None,
) -> None:
litellm.Router(
model_list=[
{
"model_name": "translate",
"litellm_params": {
"model": "openai/my-translator",
"api_key": "sk-test",
"output_cost_per_second": 0.001,
},
"model_info": {"id": "custom-translate-output-priced"},
}
]
)
public_input_rate: Final = litellm.model_cost["gpt-realtime-translate"]["input_cost_per_second"]
assert public_input_rate > 0
assert _translation_session_cost(
"custom-translate-output-priced", model="my-translator"
) == pytest.approx(60 * 0.001 + 60 * public_input_rate)
def test_realtime_translation_rate_ignores_a_session_model_the_deployment_does_not_run(
_local_model_cost_map: None,
) -> None: