chore(typing): replace Any and bare containers added in the last day

Type the annotations that landed in the last 24 hours and ratchet the lint budgets down accordingly. No behavior change.

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

litellm/integrations/otel/plumbing/routing.py

@@ -15,7 +15,7 @@ from collections import OrderedDict
 from collections.abc import Mapping
 from dataclasses import dataclass
 from types import MappingProxyType
-from typing import Any, Final, TypeAlias
+from typing import Final, TypeAlias
 from urllib.parse import quote
 
 from opentelemetry.sdk.trace import TracerProvider
@@ -32,6 +32,7 @@ from litellm.integrations.otel.presets import (
     dynamic_otlp_headers,
     project_routing_headers,
 )
+from litellm.types.utils import StandardCallbackDynamicParams
 
 # Exporter kinds that ignore headers — never rewritten with dynamic credentials.
 _NON_OTLP_KINDS: Final = ("console", "in_memory", "inmemory", "memory")
@@ -166,7 +167,7 @@ class TenantTracerCache:
     def route_for(
         self,
         default: Tracer,
-        dynamic_params: Any,
+        dynamic_params: StandardCallbackDynamicParams | None,
         auth_metadata: Mapping[str, str] | None = None,
     ) -> TenantRoute:
         """Return the tracer (and trace-detachment flag) for this request.

litellm/proxy/common_request_processing.py

@@ -4,7 +4,7 @@ import json
 import logging
 import math
 import traceback
-from collections.abc import AsyncGenerator, Awaitable, Callable, Mapping
+from collections.abc import AsyncGenerator, Awaitable, Callable, Mapping, Sequence
 from datetime import datetime
 from functools import lru_cache
 from types import MappingProxyType
@@ -279,7 +279,7 @@ def _deferred_stream_logging_is_armed(request_data: dict) -> bool:
     )
 
 
-def _assembled_model_came_from_a_later_chunk(chunks: list, assembled_model: object) -> bool:
+def _assembled_model_came_from_a_later_chunk(chunks: Sequence[object], assembled_model: object) -> bool:
     """Report whether stream_chunk_builder picked a model the first chunk did not carry.
 
     Azure Model Router puts the routed model on the chunks after the first one, and the
@@ -301,7 +301,10 @@ def _assembled_model_came_from_a_later_chunk(chunks: list, assembled_model: obje
     )
 
 
-def _assembled_model_is_the_name_the_client_asked_for(request_data: dict, assembled_model: object) -> bool:
+def _assembled_model_is_the_name_the_client_asked_for(
+    request_data: Mapping[str, object],
+    assembled_model: object,
+) -> bool:
     """Report whether the assembled model is the public name the proxy stamps onto chunks.
 
     That stamp is what leaves an unpriced alias on the partial response, so the deployment's

litellm/proxy/common_utils/reset_budget_job.py

@@ -1182,7 +1182,7 @@ class ResetBudgetJob:
             if not raw:
                 continue
             row_id: str = row[source.id_column]
-            windows: list = raw if isinstance(raw, list) else json.loads(raw)
+            windows: list[dict[str, object]] = raw if isinstance(raw, list) else json.loads(raw)
             changed = False
             for window in windows:
                 counter_key = f"{source.counter_prefix}:{row_id}:window:{window['budget_duration']}"

litellm/proxy/spend_tracking/budget_reservation.py

@@ -1261,7 +1261,7 @@ def _count_input_tokens_for_models(
 _INPUT_SIZE_FIELDS: Final = ("messages", "prompt", "input", "query", "documents", "tools", "tool_choice")
 
 
-def _approximate_input_size(request_body: dict) -> int:
+def _approximate_input_size(request_body: Mapping[str, object]) -> int:
     """Length of the request's input text, a cheap stand-in for tokenizing cost.
 
     Every field _count_input_tokens hands the tokenizer is sized here, and

ruff-strict-budget.json

@@ -24,7 +24,7 @@
     "limit": 133
   },
   "ANN401": {
-    "limit": 1188
+    "limit": 1187
   },
   "ASYNC230": {
     "limit": 11
@@ -234,7 +234,7 @@
     "limit": 5
   },
   "TID251": {
-    "limit": 1212
+    "limit": 1211
   },
   "TRY002": {
     "limit": 524

tests/test_litellm/llms/anthropic/experimental_pass_through/adapters/test_anthropic_experimental_pass_through_adapters_transformation.py

@@ -635,7 +635,7 @@ def test_translate_anthropic_to_openai_orders_top_level_and_midturn_system():
 
 
 def _translate_with_metadata(
-    model: str, metadata: dict[str, Any], custom_llm_provider: str | None
+    model: str, metadata: dict[str, str], custom_llm_provider: str | None
 ) -> dict[str, Any]:
     openai_request, _ = LiteLLMAnthropicMessagesAdapter().translate_anthropic_to_openai(
         anthropic_message_request={

tests/test_litellm/proxy/common_utils/test_reset_budget_job.py

@@ -1948,14 +1948,14 @@ class FakePodLockManager:
         if self.redis_cache is not None:
             self.redis_cache.async_get_cache = AsyncMock(return_value="another-pod" if held_by_other else None)
         self._acquired = acquired
-        self.acquire_calls: List[Dict[str, Any]] = []
+        self.acquire_calls: List[Dict[str, str | int | None]] = []
         self.release_calls: List[str] = []
 
     @staticmethod
     def get_redis_lock_key(cronjob_id: str) -> str:
         return f"cronjob_lock:{cronjob_id}"
 
-    async def acquire_lock(self, cronjob_id: str, ttl: Any = None) -> bool:
+    async def acquire_lock(self, cronjob_id: str, ttl: int | None = None) -> bool:
         self.acquire_calls.append({"cronjob_id": cronjob_id, "ttl": ttl})
         return self._acquired