fix(drop_params): honor string flags in litellm_settings and responses, and fail open on non-flag values

litellm/proxy/proxy_server.py

@@ -279,6 +279,7 @@ from litellm.litellm_core_utils.audio_utils.utils import resolve_speech_media_ty
 from litellm.litellm_core_utils.core_helpers import (
     _get_parent_otel_span_from_kwargs,
     get_litellm_metadata_from_kwargs,
+    normalize_drop_params,
 )
 from litellm.litellm_core_utils.credential_accessor import CredentialAccessor
 from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
@@ -5508,6 +5509,8 @@ class ProxyConfig:
 
                     parse_budget_reset_time(value)
                     setattr(litellm, key, value)
+                elif key == "drop_params":
+                    litellm.drop_params = bool(normalize_drop_params(value))
                 else:
                     verbose_proxy_logger.debug(
                         "%s setting litellm.%s=%s%s",

litellm/responses/main.py

@@ -17,6 +17,7 @@ from litellm.completion_extras.litellm_responses_transformation.transformation i
 from litellm.constants import request_timeout
 from litellm.integrations.anthropic_cache_control_hook import CARRY_UNMATCHED_MESSAGE_POINTS
 from litellm.litellm_core_utils.asyncify import run_async_function
+from litellm.litellm_core_utils.core_helpers import normalize_drop_params
 from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
 from litellm.litellm_core_utils.prompt_templates.common_utils import (
     update_responses_input_with_model_file_ids,
@@ -1253,7 +1254,7 @@ def responses(
             responses_api_provider_config=responses_api_provider_config,
             response_api_optional_params=response_api_optional_params,
             allowed_openai_params=allowed_openai_params,
-            drop_params=request_drop_params if isinstance(request_drop_params, bool) else None,
+            drop_params=normalize_drop_params(request_drop_params),
         )
 
         litellm_logging_obj.update_from_kwargs(
@@ -2081,7 +2082,7 @@ def compact_responses(
             responses_api_provider_config=responses_api_provider_config,
             response_api_optional_params=response_api_optional_params,
             allowed_openai_params=None,
-            drop_params=request_drop_params if isinstance(request_drop_params, bool) else None,
+            drop_params=normalize_drop_params(request_drop_params),
         )
 
         # Pre Call logging

litellm/types/router.py

@@ -408,9 +408,11 @@ class GenericLiteLLMParams(CredentialLiteLLMParams, CustomPricingLiteLLMParams):
 
     @field_validator("drop_params", mode="before")
     @classmethod
-    def coerce_drop_params(cls, value: object) -> object:
+    def coerce_drop_params(cls, value: object) -> bool | str | None:
         normalized: Final = normalize_drop_params(value)
-        return value if normalized is None else normalized
+        if normalized is not None:
+            return normalized
+        return value if isinstance(value, str) else None
 
     def __contains__(self, key) -> bool:
         # Define custom behavior for the 'in' operator

tests/test_litellm/proxy/proxy_server/test_proxy_config.py

@@ -2431,6 +2431,49 @@ def test_ProxyConfig__add_deployment_turns_stored_drop_params_string_into_bool(m
     assert deployment.litellm_params.drop_params is True
 
 
+def test_ProxyConfig__add_deployment_keeps_loading_rows_after_a_non_flag_drop_params(monkeypatch):
+    monkeypatch.setenv("LITELLM_SALT_KEY", "sk-1234")
+    fake_router = MagicMock()
+    fake_router.upsert_deployment = MagicMock(return_value=True)
+    monkeypatch.setattr("litellm.proxy.proxy_server.llm_router", fake_router)
+    pc = ProxyConfig()
+
+    def db_model(model_id, drop_params):
+        return SimpleNamespace(
+            model_id=model_id,
+            model_name="gpt-5-nano",
+            model_info={"id": model_id},
+            litellm_params={
+                "model": encrypt_value_helper(value="openai/gpt-5-nano"),
+                "drop_params": encrypt_value_helper(value=drop_params),
+            },
+            blocked=False,
+        )
+
+    added = pc._add_deployment(db_models=[db_model("bad-row", 2), db_model("good-after", "true")])
+    deployments = [call.kwargs["deployment"] for call in fake_router.upsert_deployment.call_args_list]
+
+    assert added == 2
+    assert [d.litellm_params.drop_params for d in deployments] == [None, True]
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("configured, expected", [("true", True), ("false", False)])
+async def test_ProxyConfig_load_config_turns_litellm_settings_drop_params_string_into_bool(
+    tmp_path, monkeypatch, configured, expected
+):
+    f = tmp_path / "c.yaml"
+    f.write_text(f'model_list: []\nlitellm_settings:\n  drop_params: "{configured}"\n')
+    monkeypatch.setattr("litellm.proxy.proxy_server.prisma_client", None)
+    monkeypatch.setattr("litellm.proxy.proxy_server.store_model_in_db", False)
+    monkeypatch.delenv("LITELLM_CONFIG_BUCKET_NAME", raising=False)
+    monkeypatch.setattr(litellm, "drop_params", not expected)
+
+    await ProxyConfig().load_config(router=None, config_file_path=str(f))
+
+    assert litellm.drop_params is expected
+
+
 # ---------------------------------------------------------------------------
 # ProxyConfig.decrypt_model_list_from_db
 # ---------------------------------------------------------------------------

tests/test_litellm/responses/test_responses_api_request_body.py

@@ -246,8 +246,10 @@ async def test_aresponses_keeps_include_obfuscation_in_stream_options():
 
 
 @pytest.mark.asyncio
+@pytest.mark.parametrize("drop_params", [True, "true"])
 async def test_aresponses_request_level_drop_params_drops_bedrock_mantle_service_tier(
     monkeypatch,
+    drop_params,
 ):
     """
     Request-level drop_params=True (as the proxy injects for agentic CLIs) must
@@ -271,7 +273,7 @@ async def test_aresponses_request_level_drop_params_drops_bedrock_mantle_service
             aws_region_name="us-east-1",
             input="hi",
             service_tier="priority",
-            drop_params=True,
+            drop_params=drop_params,
         )
 
         mock_post.assert_called_once()

tests/test_litellm/types/test_router.py

@@ -1,5 +1,4 @@
 import pytest
-from pydantic import ValidationError
 
 from litellm.types.router import (
     SPECIAL_MODEL_INFO_PARAMS,
@@ -109,6 +108,6 @@ def test_drop_params_coerces_flags_and_keeps_unresolved_strings(value, expected)
     assert GenericLiteLLMParams(drop_params=value).drop_params == expected
 
 
-def test_drop_params_rejects_non_flag_non_string_values():
-    with pytest.raises(ValidationError):
-        GenericLiteLLMParams(drop_params=2)
+@pytest.mark.parametrize("value", [2, 2.5, [], {}])
+def test_drop_params_ignores_non_flag_non_string_values(value):
+    assert GenericLiteLLMParams(drop_params=value).drop_params is None