mirror of
https://github.com/BerriAI/litellm.git
synced 2026-11-16 09:22:22 +00:00
fix(dashscope): align implicit cache pricing reports
This commit is contained in:
parent
cfb665c21a
commit
4faa0359cb
3 changed files with 81 additions and 6 deletions
|
|
@ -1586,6 +1586,14 @@ def calculate_prompt_caching_savings(
|
|||
threshold_is_inclusive=_uses_inclusive_token_thresholds(custom_llm_provider),
|
||||
missing_cache_read_uses_input=True,
|
||||
)
|
||||
provider_cache_read_cost: Final = _resolve_cache_read_cost_rate(
|
||||
custom_llm_provider=custom_llm_provider,
|
||||
model_info=model_info,
|
||||
usage=usage,
|
||||
current_time=billed_at,
|
||||
)
|
||||
if provider_cache_read_cost is not None:
|
||||
cache_read_cost = provider_cache_read_cost
|
||||
write_rate: Final = cache_creation_cost or prompt_base_cost
|
||||
write_rate_1h: Final = cache_creation_cost_above_1hr or write_rate
|
||||
prompt_tokens_details: Final = parse_prompt_tokens_details(usage)
|
||||
|
|
|
|||
|
|
@ -122,6 +122,20 @@ def _flat_rates(
|
|||
)
|
||||
|
||||
|
||||
def _tier_cache_read_rate(
|
||||
model_info: ModelInfo,
|
||||
tier: dict[str, object],
|
||||
cache_read_mode: QwenContextCacheMode | None,
|
||||
) -> float:
|
||||
if cache_read_mode != "implicit":
|
||||
return tier_rate(tier, "cache_read_input_token_cost", "input_cost_per_token")
|
||||
if "implicit_cache_read_input_token_cost" in tier:
|
||||
return tier_rate(tier, "implicit_cache_read_input_token_cost")
|
||||
if model_info.get("implicit_cache_read_input_token_cost") is not None:
|
||||
return _flat_cache_read_rate(model_info, cache_read_mode)
|
||||
return tier_rate(tier, "cache_read_input_token_cost", "input_cost_per_token")
|
||||
|
||||
|
||||
def _tier_rates(
|
||||
model_info: ModelInfo,
|
||||
tier: dict[str, object],
|
||||
|
|
@ -133,10 +147,10 @@ def _tier_rates(
|
|||
flat_rates: Final = _flat_rates(model_info, cache_read_mode)
|
||||
tier_declares_output: Final = "output_cost_per_token" in tier
|
||||
tier_declares_reasoning: Final = "output_cost_per_reasoning_token" in tier
|
||||
cache_read_rate: Final = (
|
||||
tier_rate(tier, "implicit_cache_read_input_token_cost")
|
||||
if cache_read_mode == "implicit" and "implicit_cache_read_input_token_cost" in tier
|
||||
else tier_rate(tier, "cache_read_input_token_cost", "input_cost_per_token")
|
||||
cache_read_rate: Final = _tier_cache_read_rate(
|
||||
model_info=model_info,
|
||||
tier=tier,
|
||||
cache_read_mode=cache_read_mode,
|
||||
)
|
||||
return TokenRates(
|
||||
input_rate=tier_rate(tier, "input_cost_per_token"),
|
||||
|
|
|
|||
|
|
@ -18,6 +18,7 @@ import pytest
|
|||
import litellm
|
||||
from litellm.litellm_core_utils.llm_cost_calc.utils import (
|
||||
_resolve_cache_read_cost_rate,
|
||||
calculate_prompt_caching_savings,
|
||||
get_token_type_cost_breakdown,
|
||||
)
|
||||
from litellm.llms.dashscope.cost_calculator import (
|
||||
|
|
@ -25,6 +26,7 @@ from litellm.llms.dashscope.cost_calculator import (
|
|||
)
|
||||
from litellm.types.utils import (
|
||||
CompletionTokensDetailsWrapper,
|
||||
ModelInfo,
|
||||
PromptTokensDetailsWrapper,
|
||||
Usage,
|
||||
)
|
||||
|
|
@ -200,12 +202,63 @@ class TestDashscopeCostCalculator:
|
|||
usage = Usage(
|
||||
prompt_tokens=1000,
|
||||
completion_tokens=0,
|
||||
prompt_tokens_details={"cached_tokens": 600, "cache_type": "ephemeral"},
|
||||
prompt_tokens_details={"cached_tokens": 600},
|
||||
)
|
||||
|
||||
prompt_cost, _ = dashscope_cost_per_token(model="qwen-tiered-cache-mode-test", usage=usage)
|
||||
|
||||
assert math.isclose(prompt_cost, (400 * 1e-06) + (600 * 1e-07), rel_tol=1e-10)
|
||||
assert math.isclose(prompt_cost, (400 * 1e-06) + (600 * 2e-07), rel_tol=1e-10)
|
||||
|
||||
def test_dashscope_tiered_pricing_uses_model_level_implicit_cache_read_rate(self):
|
||||
litellm.model_cost["dashscope/qwen-tiered-model-implicit-rate-test"] = {
|
||||
"litellm_provider": "dashscope",
|
||||
"mode": "chat",
|
||||
"implicit_cache_read_input_token_cost": 2e-07,
|
||||
"tiered_pricing": [
|
||||
{
|
||||
"range": [0, 1000],
|
||||
"input_cost_per_token": 1e-06,
|
||||
"output_cost_per_token": 4e-06,
|
||||
"cache_read_input_token_cost": 1e-07,
|
||||
}
|
||||
],
|
||||
}
|
||||
usage = Usage(
|
||||
prompt_tokens=1000,
|
||||
completion_tokens=0,
|
||||
prompt_tokens_details={"cached_tokens": 600},
|
||||
)
|
||||
|
||||
prompt_cost, _ = dashscope_cost_per_token(
|
||||
model="qwen-tiered-model-implicit-rate-test",
|
||||
usage=usage,
|
||||
)
|
||||
|
||||
assert math.isclose(prompt_cost, (400 * 1e-06) + (600 * 2e-07), rel_tol=1e-10)
|
||||
|
||||
def test_dashscope_prompt_caching_savings_uses_implicit_cache_read_rate(self):
|
||||
model_info: ModelInfo = {
|
||||
"key": "qwen-cache-savings-mode-test",
|
||||
"litellm_provider": "dashscope",
|
||||
"mode": "chat",
|
||||
"input_cost_per_token": 1e-06,
|
||||
"output_cost_per_token": 4e-06,
|
||||
"cache_read_input_token_cost": 1e-07,
|
||||
"implicit_cache_read_input_token_cost": 2e-07,
|
||||
}
|
||||
usage = Usage(
|
||||
prompt_tokens=1000,
|
||||
completion_tokens=0,
|
||||
prompt_tokens_details={"cached_tokens": 600},
|
||||
)
|
||||
|
||||
savings = calculate_prompt_caching_savings(
|
||||
model_info=model_info,
|
||||
usage=usage,
|
||||
custom_llm_provider="dashscope",
|
||||
)
|
||||
|
||||
assert math.isclose(savings, 600 * (1e-06 - 2e-07), rel_tol=1e-10)
|
||||
|
||||
@pytest.mark.parametrize(
|
||||
("cache_type", "expected_cache_rate"),
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue