Merge f7d6bb579d into ed4caebb65

litellm/litellm_core_utils/streaming_handler.py

@@ -55,6 +55,7 @@ FUNCTION_CALL_ATTRIBUTE: Final = "function_call"
 _SYNC_ITER_EXHAUSTED: Final = object()
 
 _GCHUNK_FIELDS: Final[frozenset] = frozenset(GChunk.__annotations__)
+_GCHUNK_REQUIRED_FIELDS: Final[frozenset] = frozenset(GChunk.__required_keys__)
 _USAGE_COST_HEADER_PROVIDERS: Final[frozenset[str]] = frozenset({LlmProviders.OPENROUTER.value})
 
 
@@ -2467,7 +2468,7 @@ def generic_chunk_has_all_required_fields(chunk: dict) -> bool:
     :param chunk: The dictionary to check.
     :return: True if all required fields are present, False otherwise.
     """
-    return all(key in _GCHUNK_FIELDS for key in chunk)
+    return _GCHUNK_REQUIRED_FIELDS <= chunk.keys() <= _GCHUNK_FIELDS
 
 
 def convert_generic_chunk_to_model_response_stream(

tests/unit/litellm_core_utils/test_streaming_handler.py

@@ -17,6 +17,8 @@ from litellm.litellm_core_utils.streaming_handler import (
     CustomStreamWrapper,
     _ProviderChunkEarlyReturn,
     _ProviderChunkParsed,
+    convert_generic_chunk_to_model_response_stream,
+    generic_chunk_has_all_required_fields,
 )
 from litellm.types.utils import (
     CompletionTokensDetailsWrapper,
@@ -4985,6 +4987,24 @@ async def test_async_stream_without_usage_counts_tokens_off_the_event_loop():
     assert_loop_stayed_free(took, lags)
 
 
+def test_generic_chunk_predicate_rejects_partial_chunk():
+    partial_chunk = {"is_finished": False}
+    assert generic_chunk_has_all_required_fields(partial_chunk) is False
+
+    full_chunk = {
+        "text": "hello",
+        "is_finished": True,
+        "finish_reason": "stop",
+        "usage": None,
+        "index": 0,
+    }
+    assert generic_chunk_has_all_required_fields(full_chunk) is True
+
+    stream = convert_generic_chunk_to_model_response_stream(full_chunk)
+    assert stream.choices[0].delta.content == "hello"
+    assert stream.finish_reason == "stop"
+
+
 @pytest.mark.parametrize("sync_mode", [True, False])
 @pytest.mark.asyncio
 async def test_openai_stream_relays_the_served_service_tier_on_every_chunk_including_usage(

tests/unit/litellm_core_utils/test_streaming_overhead.py

@@ -127,10 +127,11 @@ def test_gchunk_fields_is_frozenset():
 
 
 def test_generic_chunk_has_all_required_fields_uses_module_constant(monkeypatch):
-    """generic_chunk_has_all_required_fields must use _GCHUNK_FIELDS, not __annotations__.
+    """generic_chunk_has_all_required_fields must use the module constants.
 
-    The check semantics: every key in `chunk` must be a known GChunk field.
-    This identifies GChunk-shaped dicts (all keys are valid GChunk fields).
+    The check semantics: every required GChunk field must be present in `chunk`
+    and every key must be a known GChunk field. Unknown keys and missing
+    required fields both reject the chunk.
     """
     valid_chunk = _make_generic_chunk("hello")
     assert generic_chunk_has_all_required_fields(valid_chunk) is True
@@ -141,10 +142,9 @@ def test_generic_chunk_has_all_required_fields_uses_module_constant(monkeypatch)
     extra_key_chunk["unknown_extra_key"] = "value"
     assert generic_chunk_has_all_required_fields(extra_key_chunk) is False
 
-    # A dict with only known GChunk fields but fewer keys still passes because
-    # all its keys are valid (subset of GChunk fields).
+    # A dict missing required fields is rejected, even if every key it has is valid.
     partial_chunk = {"text": "hi", "is_finished": False}
-    assert generic_chunk_has_all_required_fields(partial_chunk) is True
+    assert generic_chunk_has_all_required_fields(partial_chunk) is False
 
 
 # ---------------------------------------------------------------------------