fix(guardrails): keep tool calls carried by a later choice of a packed multi-choice chunk

The rebuild's tool-call selection and its text-only fast path only looked at
choice 0 of each chunk, so a chunk that packs several choices (Gemini with
candidateCount above 1) lost a tool call carried by a later candidate, and a
chunk whose later choice had no tool calls at all made the rebuild raise.
Both now consider every choice in the chunk.

litellm/litellm_core_utils/streaming_chunk_builder_utils.py

@@ -427,7 +427,7 @@ class ChunkProcessor:
                 if not delta:
                     continue
                 choice_index = choice.get("index", 0)
-                for tool_call in delta.get("tool_calls", ()):
+                for tool_call in delta.get("tool_calls") or ():
                     if not tool_call:
                         continue
                     if isinstance(tool_call, dict):
@@ -478,7 +478,7 @@ class ChunkProcessor:
             choices = chunk["choices"]
             for choice in choices:
                 delta = choice.get("delta", {})
-                tool_calls = delta.get("tool_calls", [])
+                tool_calls = delta.get("tool_calls") or ()
                 choice_index = choice.get("index", 0)
 
                 for tool_call in tool_calls:

litellm/main.py

@@ -8749,6 +8749,39 @@ def _stamp_streaming_usage_cost(usage: Usage, response: ModelResponse, logging_o
         setattr(usage, "cost", computed_cost)
 
 
+_NON_TEXT_DELTA_FIELDS: Final = (
+    "tool_calls",
+    "function_call",
+    "reasoning_content",
+    "thinking_blocks",
+    "annotations",
+    "audio",
+    "images",
+    "provider_specific_fields",
+)
+
+
+def _stream_choice_delta(choice: object) -> Mapping[str, object]:
+    delta: Final = choice.get("delta", {}) if isinstance(choice, dict) else getattr(choice, "delta", {})
+    if isinstance(delta, Mapping):
+        return delta
+    if isinstance(delta, BaseModel):
+        return delta.model_dump()
+    return {}
+
+
+def _delta_carries_more_than_text(delta: Mapping[str, object]) -> bool:
+    return any(delta.get(field) is not None for field in _NON_TEXT_DELTA_FIELDS)
+
+
+def _simple_text_part(choices: Sequence[object]) -> str | None:
+    deltas: Final = tuple(_stream_choice_delta(choice) for choice in choices)
+    if any(_delta_carries_more_than_text(delta) for delta in deltas):
+        return None
+    content: Final = deltas[0].get("content")
+    return content if isinstance(content, str) else ""
+
+
 def stream_chunk_builder(
     chunks: list,
     messages: Sequence | None = None,
@@ -8793,31 +8826,11 @@ def stream_chunk_builder(
             if not chunk.get("choices"):
                 continue
 
-            choice = chunk["choices"][0]
-            delta_obj = choice.get("delta", {}) if isinstance(choice, dict) else getattr(choice, "delta", {})
-            if isinstance(delta_obj, dict):
-                delta = delta_obj
-            elif hasattr(delta_obj, "model_dump"):
-                delta = cast(dict[str, Any], delta_obj.model_dump())
-            else:
-                delta = {}
-
-            if (
-                delta.get("tool_calls") is not None
-                or delta.get("function_call") is not None
-                or delta.get("reasoning_content") is not None
-                or delta.get("thinking_blocks") is not None
-                or delta.get("annotations") is not None
-                or delta.get("audio") is not None
-                or delta.get("images") is not None
-                or delta.get("provider_specific_fields") is not None
-            ):
+            if (part := _simple_text_part(chunk["choices"])) is None:
                 is_simple_text_stream = False
                 break
-
-            content = delta.get("content")
-            if isinstance(content, str) and content:
-                simple_content_parts.append(content)
+            if part:
+                simple_content_parts.append(part)
 
         if is_simple_text_stream:
             if simple_content_parts:
@@ -8854,9 +8867,10 @@ def stream_chunk_builder(
         tool_call_chunks: Final = [
             chunk
             for chunk in chunks
-            if chunk.get("choices")
-            and "tool_calls" in chunk["choices"][0]["delta"]
-            and chunk["choices"][0]["delta"]["tool_calls"] is not None
+            if any(
+                "tool_calls" in choice["delta"] and choice["delta"]["tool_calls"] is not None
+                for choice in chunk.get("choices") or ()
+            )
         ]
 
         if len(tool_call_chunks) > 0:

tests/test_litellm/test_main.py

@@ -1659,6 +1659,68 @@ async def test_async_mock_delay():
     assert delay >= 0.01
 
 
+def test_stream_chunk_builder_keeps_tool_calls_carried_only_by_a_later_choice_of_a_multi_choice_chunk():
+    from litellm import stream_chunk_builder
+    from litellm.types.utils import (
+        ChatCompletionDeltaToolCall,
+        Delta,
+        Function,
+        ModelResponseStream,
+        StreamingChoices,
+    )
+
+    def chunk(choices: list[StreamingChoices]) -> ModelResponseStream:
+        return ModelResponseStream(
+            id="chatcmpl-multi-choice",
+            created=1751934860,
+            model="gpt-4.1-mini",
+            object="chat.completion.chunk",
+            choices=choices,
+        )
+
+    chunks = [
+        chunk(
+            [
+                StreamingChoices(index=0, delta=Delta(role="assistant", content="hello")),
+                StreamingChoices(
+                    index=1,
+                    delta=Delta(
+                        role="assistant",
+                        tool_calls=[
+                            ChatCompletionDeltaToolCall(
+                                id="call_1",
+                                index=0,
+                                type="function",
+                                function=Function(name="lookup_fruit", arguments='{"fruit":'),
+                            )
+                        ],
+                    ),
+                ),
+            ]
+        ),
+        chunk(
+            [
+                StreamingChoices(index=0, delta=Delta(content=" world"), finish_reason="stop"),
+                StreamingChoices(
+                    index=1,
+                    delta=Delta(
+                        tool_calls=[ChatCompletionDeltaToolCall(index=0, function=Function(arguments='"kiwi"}'))]
+                    ),
+                    finish_reason="tool_calls",
+                ),
+            ]
+        ),
+    ]
+
+    response = stream_chunk_builder(chunks=chunks)
+
+    tool_calls = response.choices[0].message.tool_calls
+    assert tool_calls is not None
+    assert [(call.id, call.function.name, call.function.arguments) for call in tool_calls] == [
+        ("call_1", "lookup_fruit", '{"fruit":"kiwi"}')
+    ]
+
+
 def test_stream_chunk_builder_thinking_blocks():
     from litellm import stream_chunk_builder
     from litellm.types.utils import Delta, ModelResponseStream, StreamingChoices