Merge 2d0ee09e58 into 0c98afa780

litellm/litellm_core_utils/litellm_logging.py

@@ -4874,7 +4874,7 @@ def _maybe_construct_otel_v2(callback_name: str, _in_memory_loggers: list[Custom
     for callback in _in_memory_loggers:
         if (
             isinstance(callback, OpenTelemetryV2)
-            and getattr(callback, "callback_name", None) == callback_name
+            and callback.callback_name == callback_name
             and (serves_a_destination or not _exports_nowhere(callback.config))
         ):
             return callback

litellm/litellm_core_utils/prompt_templates/common_utils.py

@@ -2014,11 +2014,11 @@ def strip_encrypted_reasoning_from_messages(messages: object) -> None:
     """
     if not isinstance(messages, list):
         return
-    for content in _anthropic_content_lists(cast(list[object], messages)):  # cast-ok: untyped client json
+    for content in anthropic_content_lists(cast(list[object], messages)):  # cast-ok: untyped client json
         _strip_encrypted_reasoning_from_blocks(content)
 
 
-def _anthropic_content_lists(messages: Sequence[object]) -> Iterator[object]:
+def anthropic_content_lists(messages: Sequence[object]) -> Iterator[object]:
     return (
         cast(list[object], content)  # cast-ok: narrowed by isinstance
         for message in messages

litellm/litellm_core_utils/streaming_handler.py

@@ -1482,7 +1482,7 @@ class CustomStreamWrapper:
                 "is_finished": chunk_finish_reason is not None,
                 "finish_reason": chunk_finish_reason,
                 "original_chunk": cached_chunk,
-                "tool_calls": (getattr(cached_choice.delta, "tool_calls", None) if cached_choice is not None else None),
+                "tool_calls": cached_choice.delta.tool_calls if cached_choice is not None else None,
             }
 
             completion_obj["content"] = response_obj["text"]

litellm/llms/anthropic/experimental_pass_through/adapters/streaming_iterator.py

@@ -11,7 +11,6 @@ from typing import (
     Final,
     Literal,
     Protocol,
-    cast,  # noqa: TID251  # rebuilt message_delta dict spans the ContentBlockDelta/MessageBlockDelta union
     get_args,
 )
 
@@ -25,6 +24,7 @@ from litellm.types.llms.anthropic import (
     ContentBlockDelta,
     ContextManagementResponse,
     MessageBlockDelta,
+    MessageDelta,
     StreamingContentBlockDeltaType,
     UsageDelta,
     UsageIteration,
@@ -1006,26 +1006,28 @@ class AnthropicStreamWrapper(AdapterCompletionStreamWrapper):
         self,
         processed_chunk: ContentBlockDelta | MessageBlockDelta,
     ) -> ContentBlockDelta | MessageBlockDelta:
-        if processed_chunk.get("type") != "message_delta" or not self._refusal_text:
+        if processed_chunk["type"] != "message_delta" or not self._refusal_text:
             return processed_chunk
-        delta: Final = cast(Mapping[str, object], processed_chunk["delta"])  # cast-ok: keys checked before use
+        delta: Final = processed_chunk["delta"]
         if delta.get("stop_reason") == "max_tokens":
             return processed_chunk
         from litellm.llms.anthropic.experimental_pass_through.messages.utils import (
             refusal_stop_details,
         )
 
-        return cast(  # cast-ok: rebuilt dict matches the message_delta TypedDict shape for this branch
-            ContentBlockDelta | MessageBlockDelta,
-            {  # mutable-ok: fresh translation payload; never mutated after construction
-                **processed_chunk,
-                "delta": {  # mutable-ok: fresh message_delta payload; never mutated after construction
-                    **delta,
-                    "stop_reason": "refusal",
-                    "stop_details": refusal_stop_details(self._refusal_text),
-                },
-            },
-        )
+        refusal_delta: Final[MessageDelta] = {  # mutable-ok: fresh message_delta payload
+            **delta,
+            "stop_reason": "refusal",
+            "stop_details": refusal_stop_details(self._refusal_text),
+        }
+        if "context_management" in processed_chunk:
+            return MessageBlockDelta(
+                type="message_delta",
+                delta=refusal_delta,
+                usage=processed_chunk["usage"],
+                context_management=processed_chunk["context_management"],
+            )
+        return MessageBlockDelta(type="message_delta", delta=refusal_delta, usage=processed_chunk["usage"])
 
     @staticmethod
     def _delta_has_content(processed_chunk: Mapping[str, object]) -> bool:

litellm/llms/anthropic/experimental_pass_through/messages/utils.py

@@ -37,7 +37,7 @@ def _mapping_field(container: object, key: str) -> object | None:
     """One key of a raw provider payload, or None when the payload is not a mapping."""
     if not isinstance(container, Mapping):
         return None
-    return cast(Mapping[str, object], container).get(key)  # cast-ok: raw payload, callers re-check every value
+    return container.get(key)
 
 
 def _mapping_str_field(container: object, key: str) -> str | None:

litellm/llms/anthropic/experimental_pass_through/responses_adapters/transformation.py

@@ -171,7 +171,7 @@ class LiteLLMAnthropicToResponsesAPIAdapter:
         cls,
         summary: Iterable[object],
         encrypted_content: object,
-    ) -> dict[str, Any] | None:  # mutable-ok: API message payload
+    ) -> dict[str, object] | None:  # mutable-ok: API message payload
         """The one Anthropic block for a Responses reasoning item.
 
         The item's encrypted reasoning rides the block's opaque field (`signature`, or
@@ -200,7 +200,7 @@ class LiteLLMAnthropicToResponsesAPIAdapter:
     @classmethod
     def _assistant_group_to_input_items(
         cls, group: tuple[Mapping[str, object], ...]
-    ) -> tuple[dict[str, Any], ...]:  # mutable-ok: API message payload
+    ) -> tuple[dict[str, object], ...]:  # mutable-ok: API message payload
         first: Final = group[0]
         btype: Final = first.get("type")
         if btype in ("thinking", "redacted_thinking"):

litellm/proxy/_experimental/mcp_server/tool_search.py

@@ -119,14 +119,15 @@ def with_mcp_proxy_identity(tool: Tool, server_id: str) -> Tool:
 
 
 def _mcp_proxy_identity(tool: Tool) -> MCPProxyToolIdentity:
-    identity: Final = (tool.meta or {}).get(_MCP_PROXY_IDENTITY_META_KEY)  # mutable-ok: absent metadata default
+    identity: Final = None if tool.meta is None else tool.meta.get(_MCP_PROXY_IDENTITY_META_KEY)
     if not isinstance(identity, Mapping):
         raise TypeError("MCP proxy tool identity is missing")
     server_id: Final = identity.get("server_id")
     tool_name: Final = identity.get("tool_name")
     if not isinstance(server_id, str) or not isinstance(tool_name, str):
         raise TypeError("MCP proxy tool identity is invalid")
-    return {"server_id": server_id, "tool_name": tool_name}  # mutable-ok: TypedDict identity payload
+    resolved: Final[MCPProxyToolIdentity] = {"server_id": server_id, "tool_name": tool_name}
+    return resolved
 
 
 def mcp_proxy_tool_id(tool: Tool) -> str:

litellm/proxy/pass_through_endpoints/llm_provider_handlers/vertex_passthrough_logging_handler.py

@@ -1,5 +1,6 @@
 import asyncio
 import re
+from collections.abc import Mapping
 from datetime import datetime
 from typing import TYPE_CHECKING, Any, Final, cast
 from urllib.parse import urlparse
@@ -380,7 +381,7 @@ class VertexPassthroughLoggingHandler:
     @staticmethod
     def _is_audio_predict_response(
         model: str,
-        json_response: dict,  # mutable-ok: predicate inspects the decoded provider response dictionary without mutation
+        json_response: Mapping[str, object],
     ) -> bool:
         return (
             VertexPassthroughLoggingHandler._get_audio_prediction_count(json_response=json_response) > 0
@@ -389,7 +390,7 @@ class VertexPassthroughLoggingHandler:
 
     @staticmethod
     def _get_audio_prediction_count(
-        json_response: dict,  # mutable-ok: counter inspects the decoded provider response dictionary without mutation
+        json_response: Mapping[str, object],
     ) -> int:
         predictions: Final = json_response.get("predictions")
         if not isinstance(predictions, list):

litellm/router_utils/pre_call_checks/encrypted_content_affinity_check.py

@@ -49,6 +49,7 @@ from litellm.exceptions import (
 )
 from litellm.integrations.custom_logger import CustomLogger, Span
 from litellm.litellm_core_utils.prompt_templates.common_utils import (
+    anthropic_content_lists,
     encrypted_content_of_block,
     strip_encrypted_reasoning_from_messages,
 )
@@ -154,10 +155,7 @@ class EncryptedContentAffinityCheck(CustomLogger):
             return iter(())
         return (
             cast(Mapping[str, object], block)  # cast-ok: narrowed by isinstance
-            for message in cast(list[object], messages)  # cast-ok: narrowed by isinstance
-            if isinstance(message, Mapping)
-            for content in (cast(Mapping[str, object], message).get("content"),)  # cast-ok: narrowed by isinstance
-            if isinstance(content, list)
+            for content in anthropic_content_lists(cast(list[object], messages))  # cast-ok: narrowed by isinstance
             for block in cast(list[object], content)  # cast-ok: narrowed by isinstance
             if isinstance(block, Mapping)
         )

litellm/types/llms/bedrock.py

@@ -557,7 +557,6 @@ class AmazonTitanMultimodalEmbeddingResponse(TypedDict):
     message: str  # Specifies any errors that occur during generation.
 
 
-# TwelveLabs Marengo Embed types
 TWELVELABS_EMBEDDING_INPUT_TYPES = Literal["text", "image", "video", "audio"]
 TWELVELABS_EMBEDDING_OPTIONS = Literal["visual-text", "visual-image", "audio"]