scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-10-29 06:33:28 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 267

fix(pricing): preserve cache-only registered price overrides

Browse source
This commit is contained in:
Joshua Valluru 2026-10-19 16:59:08 -07:00
parent 4744f4b621
commit 5cb159663e
2 changed files with 36 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

8
litellm/cost_calculator.py
View file

@ -137,6 +137,7 @@ from litellm.types.utils import (
TranscriptionUsageTokensObject,
Usage,
VectorStoreSearchResponse,
is_server_derived_pricing_key,
)
from litellm.utils import (
CallTypes,
@ -856,7 +857,12 @@ _NON_TOKEN_RATE_FIELDS: Final = frozenset(
def _cost_map_entry_prices_anything(entry: Mapping[str, object]) -> bool:
return any(
value is not None and (field in _NON_TOKEN_RATE_FIELDS or ("cost_per" in field and "token" in field))
value is not None
and (
field in _NON_TOKEN_RATE_FIELDS
or (field.startswith("cache_") and is_server_derived_pricing_key(field))
or ("cost_per" in field and "token" in field)
)
for field, value in entry.items()
)

29
tests/unit/test_cost_calculator.py
View file

@ -9917,3 +9917,32 @@ def test_bedrock_converse_profile_ignores_registered_capabilities_without_prices
custom_llm_provider="bedrock",
optional_params={"model_id": f"us.{model}"},
) == pytest.approx(10 * prices["input_cost_per_token"] + 5 * prices["output_cost_per_token"])
@pytest.mark.parametrize("rate", (0.0, 0.001))
@pytest.mark.parametrize("cache_field", ("cache_read_input_token_cost", "cache_creation_input_token_cost"))
def test_bedrock_converse_profile_preserves_cache_only_registered_prices(
_local_model_cost_map: None, rate: float, cache_field: str
) -> None:
model: Final = "openai.gpt-6-sol"
litellm.register_model({model: {cache_field: rate}}, persist_across_reloads=False)
response: Final = ModelResponse(
model=model,
usage=Usage(
prompt_tokens=5,
completion_tokens=0,
total_tokens=5,
prompt_tokens_details=PromptTokensDetailsWrapper(
cached_tokens=5 if cache_field == "cache_read_input_token_cost" else 0
),
cache_creation_input_tokens=5 if cache_field == "cache_creation_input_token_cost" else 0,
),
)
expected: Final = completion_cost(completion_response=response, model=model, custom_llm_provider="bedrock")
assert expected == pytest.approx(5 * rate)
assert completion_cost(
completion_response=response,
model=model,
custom_llm_provider="bedrock",
optional_params={"model_id": f"us.{model}"},
) == pytest.approx(expected)