mirror of
https://github.com/BerriAI/litellm.git
synced 2026-11-17 09:34:48 +00:00
fix(azure_ai): bill Foundry Cohere Embed V5 text and image tokens at their own rates
This commit is contained in:
parent
0070ce23b3
commit
6177bcc25b
2 changed files with 35 additions and 0 deletions
|
|
@ -57,6 +57,7 @@ class _CohereBilledUnits(BaseModel):
|
|||
model_config = ConfigDict(frozen=True, hide_input_in_errors=True)
|
||||
|
||||
input_tokens: int | None = None
|
||||
text_tokens: int | None = None
|
||||
image_tokens: int | None = None
|
||||
images: int | None = None
|
||||
|
||||
|
|
@ -116,6 +117,14 @@ def _billed_usage(billed_units: _CohereBilledUnits, image_billing: CohereImageTo
|
|||
image_count=billed_units.images,
|
||||
),
|
||||
)
|
||||
case CohereImageTokenBilling.FOLDED_INTO_INPUT_TOKENS if billed_units.text_tokens is not None:
|
||||
return _prompt_usage(
|
||||
billed_units.text_tokens + (billed_units.image_tokens or 0),
|
||||
PromptTokensDetailsWrapper(
|
||||
text_tokens=billed_units.text_tokens,
|
||||
image_tokens=billed_units.image_tokens or None,
|
||||
),
|
||||
)
|
||||
case CohereImageTokenBilling.FOLDED_INTO_INPUT_TOKENS:
|
||||
return _prompt_usage(billed_units.input_tokens or 0, None)
|
||||
|
||||
|
|
|
|||
|
|
@ -604,6 +604,32 @@ def test_native_cohere_embed_relay_prices_an_image_at_the_input_tokens_foundry_a
|
|||
|
||||
|
||||
|
||||
@pytest.mark.parametrize("model", ["Cohere-Embed-V5-Pro", "Cohere-Embed-V5-Fast"])
|
||||
@pytest.mark.parametrize(
|
||||
("input_tokens", "text_tokens", "image_tokens"),
|
||||
[(1017, 6, 258), (24, 0, 6)],
|
||||
ids=["text_and_image", "image_only"],
|
||||
)
|
||||
def test_native_cohere_embed_relay_prices_itemized_text_and_image_tokens_at_their_own_rates(
|
||||
model: str, input_tokens: int, text_tokens: int, image_tokens: int
|
||||
) -> None:
|
||||
body: Final = {
|
||||
**COHERE_EMBED_BODY,
|
||||
"meta": {
|
||||
"api_version": {"version": "2"},
|
||||
"billed_units": {"input_tokens": input_tokens, "image_tokens": image_tokens, "text_tokens": text_tokens},
|
||||
},
|
||||
}
|
||||
result, logging_obj = _relay_logging_result(AzureAIPassthroughConfig(), model, "providers/cohere/v2/embed", body)
|
||||
model_info: Final = litellm.get_model_info(f"azure_ai/{model}")
|
||||
|
||||
assert isinstance(result, EmbeddingResponse)
|
||||
assert result.usage.prompt_tokens == text_tokens + image_tokens
|
||||
assert logging_obj.response_cost_calculator(result=result) == pytest.approx(
|
||||
text_tokens * model_info["input_cost_per_token"] + image_tokens * model_info["input_cost_per_image_token"]
|
||||
)
|
||||
|
||||
|
||||
def test_native_cohere_embed_relay_rows_keep_their_embedding_type() -> None:
|
||||
result, _ = _cohere_embed_relay_result({**COHERE_EMBED_BODY, "embeddings": {"float": TWO_VECTORS, "int8": [[1, 2]]}})
|
||||
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue