diff --git a/litellm/proxy/hooks/model_max_budget_limiter.py b/litellm/proxy/hooks/model_max_budget_limiter.py index 2a3e87f81f1..ad08bac7d84 100644 --- a/litellm/proxy/hooks/model_max_budget_limiter.py +++ b/litellm/proxy/hooks/model_max_budget_limiter.py @@ -9,6 +9,7 @@ import litellm from litellm._logging import verbose_proxy_logger from litellm.caching.caching import DualCache from litellm.integrations.custom_logger import Span +from litellm.litellm_core_utils.core_helpers import get_litellm_metadata_from_kwargs from litellm.litellm_core_utils.duration_parser import duration_in_seconds from litellm.llms.bedrock.common_utils import get_bedrock_base_model from litellm.proxy._types import Litellm_EntityType, UserAPIKeyAuth @@ -465,7 +466,7 @@ class _PROXY_VirtualKeyModelMaxBudgetLimiter(RouterBudgetLimiting): return _litellm_params: Final[dict] = kwargs.get("litellm_params", {}) or {} - _metadata: Final[dict] = _litellm_params.get("metadata", {}) or {} + _metadata: Final[dict] = get_litellm_metadata_from_kwargs(kwargs) or {} payload_metadata: Final = standard_logging_payload.get("metadata") or {} model: Final = standard_logging_payload.get("model_group") or standard_logging_payload.get("model") diff --git a/tests/test_litellm/proxy/hooks/test_max_budget_limiter.py b/tests/test_litellm/proxy/hooks/test_max_budget_limiter.py index 9fdc2e5875e..f3db79b7ea2 100644 --- a/tests/test_litellm/proxy/hooks/test_max_budget_limiter.py +++ b/tests/test_litellm/proxy/hooks/test_max_budget_limiter.py @@ -271,6 +271,40 @@ async def test_async_log_success_event_uses_requested_model_budget(): assert cache.in_memory_cache.cache_dict["virtual_key_spend:hash-1:claude-haiku-4-5:1h"] == 5.5e-05 +@pytest.mark.asyncio +async def test_async_log_success_event_reads_requested_model_from_litellm_metadata(): + cache = DualCache() + handler = _PROXY_VirtualKeyModelMaxBudgetLimiter(dual_cache=cache) + + await handler.async_log_success_event( + kwargs={ + "standard_logging_object": { + "model_group": "claude-haiku-4-5-20251001", + "model": "anthropic/claude-haiku-4-5-20251001", + "response_cost": 5.5e-05, + "metadata": {"user_api_key_hash": "hash-1"}, + }, + "litellm_params": { + "metadata": {"trace": "x"}, + "litellm_metadata": { + "user_api_key_model_max_budget": { + "claude-haiku-4-5": { + "budget_limit": 1e-4, + "time_period": "1h", + } + }, + "litellm_client_requested_model": "claude-haiku-4-5", + }, + }, + }, + response_obj=None, + start_time=0, + end_time=0, + ) + + assert cache.in_memory_cache.cache_dict["virtual_key_spend:hash-1:claude-haiku-4-5:1h"] == 5.5e-05 + + @pytest.mark.asyncio async def test_get_fallback_model_within_budget_uses_alias_budget_fallbacks(): handler = _PROXY_VirtualKeyModelMaxBudgetLimiter(dual_cache=DualCache())