fix: drop Final from loop locals to stay within the basedpyright budget

litellm/proxy/management_endpoints/key_management_endpoints.py

@@ -3555,7 +3555,7 @@ async def _attach_budget_limits_usage(key_info: dict, api_key_hash: str) -> None
         return
 
     for idx, window in enumerate(budget_limits):
-        w: Final = _budget_window_to_dict(window)
+        w = _budget_window_to_dict(window)
         if not w:
             continue
         budget_limits[idx] = w
@@ -3567,8 +3567,8 @@ async def _attach_budget_limits_usage(key_info: dict, api_key_hash: str) -> None
             max_budget = float(max_budget) if max_budget is not None else None
         except (TypeError, ValueError):
             max_budget = None
-        counter_key: Final = f"spend:key:{api_key_hash}:window:{duration}"
-        spend: Final = await get_current_spend(
+        counter_key = f"spend:key:{api_key_hash}:window:{duration}"
+        spend = await get_current_spend(
             counter_key=counter_key,
             fallback_spend=0.0,
             max_budget=max_budget,

ui/litellm-dashboard/src/lib/http/schema.d.ts

@@ -7020,6 +7020,9 @@ export interface paths {
          *         - model_max_budget: dict - Per-model budgets, e.g. {"gpt-4": {"budget_limit": 0.0005, "time_period": "30d"}}
          *         - model_max_budget_usage: dict | None - Current-window spend per model, present only when
          *           the key has per-model budgets
+         *         - budget_limits: list | None - Concurrent budget windows. Each entry includes
+         *           current_spend: spend accumulated in the window so far (read from the same cross-pod
+         *           spend counter the budget enforcement uses)
          *         - models: list - Model_name's the key is allowed to call
          *         - tpm_limit / rpm_limit: int | None - Tokens and requests per minute limits
          *         - metadata: dict - Metadata for the key, e.g. {"team": "core-infra"}