diff --git a/litellm/responses/litellm_completion_transformation/streaming_iterator.py b/litellm/responses/litellm_completion_transformation/streaming_iterator.py index ba2d595c523..b941fcbe0ea 100644 --- a/litellm/responses/litellm_completion_transformation/streaming_iterator.py +++ b/litellm/responses/litellm_completion_transformation/streaming_iterator.py @@ -1100,8 +1100,8 @@ class LiteLLMCompletionStreamingIterator(ResponsesAPIStreamingIterator): try: built = stream_chunk_builder(chunks=chunks) if built and built.choices and hasattr(built.choices[0], "message"): - self._cached_encrypted_reasoning_content = ( - LiteLLMCompletionResponsesConfig._encode_thinking_blocks(built.choices[0].message) + self._cached_encrypted_reasoning_content = LiteLLMCompletionResponsesConfig._encode_thinking_blocks( + built.choices[0].message ) return self._cached_encrypted_reasoning_content except Exception: diff --git a/tests/unit/responses/litellm_completion_transformation/test_streaming_iterator_transformation.py b/tests/unit/responses/litellm_completion_transformation/test_streaming_iterator_transformation.py index 4ef9d45b4a2..5b372924667 100644 --- a/tests/unit/responses/litellm_completion_transformation/test_streaming_iterator_transformation.py +++ b/tests/unit/responses/litellm_completion_transformation/test_streaming_iterator_transformation.py @@ -35,9 +35,7 @@ from litellm.types.utils import ( ) CHAT_COMPLETION_ID = "chatcmpl-77d33d09-effa-4cd2-9c0d-c742d4358256" -RESPONSE_ID_EVENT_TYPES = frozenset( - {"response.created", "response.in_progress", "response.completed"} -) +RESPONSE_ID_EVENT_TYPES = frozenset({"response.created", "response.in_progress", "response.completed"}) def _chunk(content: str, finish_reason: str | None = None) -> ModelResponseStream: @@ -90,11 +88,7 @@ def _build_iterator(chunks) -> LiteLLMCompletionStreamingIterator: def _response_ids(events) -> list[str]: - return [ - event.response.id - for event in events - if getattr(event, "type", None) in RESPONSE_ID_EVENT_TYPES - ] + return [event.response.id for event in events if getattr(event, "type", None) in RESPONSE_ID_EVENT_TYPES] def test_tool_call_delta_is_emitted_as_responses_events(): @@ -225,9 +219,7 @@ async def test_web_search_stream_preserves_hosted_and_client_calls(sync_mode, to content=None, provider_specific_fields={ "web_search_results": [first_result], - "web_search_calls": [ - build_web_search_call(call_ids[0], {"query": "one"}, first_result) - ] + "web_search_calls": [build_web_search_call(call_ids[0], {"query": "one"}, first_result)] if tool_type != "function" and first_result["type"] == "web_search_tool_result" else [], }, @@ -326,9 +318,7 @@ async def test_web_search_stream_preserves_hosted_and_client_calls(sync_mode, to assert {event["output_index"] for event in search_events} == {completed_indexes[item["id"]]} assert search_events[0]["item"]["status"] == "in_progress" assert search_events[-1]["item"] == item - assert item["status"] == ( - "failed" if result_kind == "error" and call_id.endswith("01Search") else "completed" - ) + assert item["status"] == ("failed" if result_kind == "error" and call_id.endswith("01Search") else "completed") assert item["action"]["type"] == "search" assert item["action"]["query"] == ("one" if call_id.endswith("01Search") else "two") assert item["action"]["queries"] == [item["action"]["query"]] @@ -437,9 +427,7 @@ def test_tool_call_arguments_are_chunked_to_match_openai_behavior(): ) # Create a chunk with a large arguments string that should be split - large_arguments = ( - '{"param1": "value1", "param2": "value2", "param3": "value3"}' # 67 chars - ) + large_arguments = '{"param1": "value1", "param2": "value2", "param3": "value3"}' # 67 chars chunk = ModelResponseStream( id="chunk-1", created=123, @@ -494,9 +482,7 @@ def test_tool_call_arguments_are_chunked_to_match_openai_behavior(): # Keep draining pending events (expected: ceil(67 / 10) = 7 delta events) while iterator._pending_tool_events: - evt = iterator._transform_chat_completion_chunk_to_response_api_chunk( - empty_chunk - ) + evt = iterator._transform_chat_completion_chunk_to_response_api_chunk(empty_chunk) if evt and evt.type == ResponsesAPIStreamEvents.FUNCTION_CALL_ARGUMENTS_DELTA: delta_events.append(evt) @@ -549,20 +535,12 @@ def test_tool_call_delta_without_id_uses_index_mapping(): while iterator._pending_tool_events: all_events.append(iterator._pending_tool_events.pop(0)) - delta_events = [ - evt - for evt in all_events - if evt.type == ResponsesAPIStreamEvents.FUNCTION_CALL_ARGUMENTS_DELTA - ] + delta_events = [evt for evt in all_events if evt.type == ResponsesAPIStreamEvents.FUNCTION_CALL_ARGUMENTS_DELTA] streamed_arguments = "".join(evt.delta for evt in delta_events) assert streamed_arguments == '{"location": "New York"}' - output_item_added_events = [ - evt - for evt in all_events - if evt.type == ResponsesAPIStreamEvents.OUTPUT_ITEM_ADDED - ] + output_item_added_events = [evt for evt in all_events if evt.type == ResponsesAPIStreamEvents.OUTPUT_ITEM_ADDED] assert len(output_item_added_events) == 1 assert output_item_added_events[0].item.id == "fc_call_abc123" assert output_item_added_events[0].item.call_id == "call_abc123" @@ -603,18 +581,10 @@ def test_parallel_tool_calls_without_ids_use_index_mapping(): while iterator._pending_tool_events: all_events.append(iterator._pending_tool_events.pop(0)) - output_item_added_events = [ - evt - for evt in all_events - if evt.type == ResponsesAPIStreamEvents.OUTPUT_ITEM_ADDED - ] + output_item_added_events = [evt for evt in all_events if evt.type == ResponsesAPIStreamEvents.OUTPUT_ITEM_ADDED] assert len(output_item_added_events) == 2 - delta_events = [ - evt - for evt in all_events - if evt.type == ResponsesAPIStreamEvents.FUNCTION_CALL_ARGUMENTS_DELTA - ] + delta_events = [evt for evt in all_events if evt.type == ResponsesAPIStreamEvents.FUNCTION_CALL_ARGUMENTS_DELTA] arguments_by_call_id = {} for evt in delta_events: arguments_by_call_id.setdefault(evt.item_id, "") @@ -667,11 +637,7 @@ def test_reused_index_with_new_call_id_marks_fallback_ambiguous(): while iterator._pending_tool_events: all_events.append(iterator._pending_tool_events.pop(0)) - delta_events = [ - evt - for evt in all_events - if evt.type == ResponsesAPIStreamEvents.FUNCTION_CALL_ARGUMENTS_DELTA - ] + delta_events = [evt for evt in all_events if evt.type == ResponsesAPIStreamEvents.FUNCTION_CALL_ARGUMENTS_DELTA] arguments_by_call_id = {} for evt in delta_events: arguments_by_call_id.setdefault(evt.item_id, "") @@ -711,22 +677,17 @@ def test_sync_streaming_events_share_the_chat_completion_response_id(): assert len(response_ids) == 3 assert len(set(response_ids)) == 1 assert ( - ResponsesAPIRequestUtils.decode_responses_api_response_id(response_ids[0])["response_id"] - == CHAT_COMPLETION_ID + ResponsesAPIRequestUtils.decode_responses_api_response_id(response_ids[0])["response_id"] == CHAT_COMPLETION_ID ) @pytest.mark.asyncio async def test_streaming_emits_every_chunk_after_priming_the_response_id(): - iterator = _build_iterator( - [_chunk("Hel"), _chunk("lo"), _chunk("!", finish_reason="stop")] - ) + iterator = _build_iterator([_chunk("Hel"), _chunk("lo"), _chunk("!", finish_reason="stop")]) events = [event async for event in iterator] - deltas = "".join( - event.delta for event in events if getattr(event, "type", None) == "response.output_text.delta" - ) + deltas = "".join(event.delta for event in events if getattr(event, "type", None) == "response.output_text.delta") assert deltas == "Hello!" @@ -772,7 +733,9 @@ async def test_leading_empty_choices_chunk_does_not_kill_the_stream(): event_types = [getattr(event, "type", None) for event in events] assert event_types.count(ResponsesAPIStreamEvents.OUTPUT_ITEM_ADDED) == 1 - assert "".join(event.delta for event in events if event.type == ResponsesAPIStreamEvents.OUTPUT_TEXT_DELTA) == "Hello!" + assert ( + "".join(event.delta for event in events if event.type == ResponsesAPIStreamEvents.OUTPUT_TEXT_DELTA) == "Hello!" + ) assert event_types[-1] == ResponsesAPIStreamEvents.RESPONSE_COMPLETED @@ -956,9 +919,7 @@ def test_streamed_unrecognized_tool_choice_is_echoed_as_auto() -> None: litellm_metadata={}, ) - response_events: Final = [ - event for event in iterator if getattr(event, "type", None) in RESPONSE_ID_EVENT_TYPES - ] + response_events: Final = [event for event in iterator if getattr(event, "type", None) in RESPONSE_ID_EVENT_TYPES] assert [event.response.tool_choice for event in response_events] == ["auto", "auto", "auto"] @@ -999,9 +960,7 @@ def _signed_thinking_and_text_in_one_chunk() -> tuple[ModelResponseStream, ...]: chunk: Final = _chunk("Hello!", finish_reason="stop") signed_block: Final = {"type": "thinking", "thinking": "", "signature": "sig_only"} delta: Final = Delta(**{**chunk.choices[0].delta.model_dump(), "thinking_blocks": [signed_block]}) - return ( - chunk.model_copy(update={"choices": [StreamingChoices(index=0, delta=delta, finish_reason="stop")]}), - ) + return (chunk.model_copy(update={"choices": [StreamingChoices(index=0, delta=delta, finish_reason="stop")]}),) StreamChunks = Callable[[], tuple[ModelResponseStream, ...]] @@ -1421,9 +1380,7 @@ async def test_anthropic_tool_search_call_streams_as_tool_search_call_item(sync_ model="claude-sonnet-5-5", object="chat.completion.chunk", choices=[ - StreamingChoices( - index=0, delta=delta, finish_reason="tool_calls" if index == len(deltas) - 1 else None - ) + StreamingChoices(index=0, delta=delta, finish_reason="tool_calls" if index == len(deltas) - 1 else None) ], ) for index, delta in enumerate(deltas) @@ -1551,10 +1508,15 @@ async def test_reasoning_output_item_done_carries_encrypted_content_and_status(s assert len(reasoning_done_events) == 1 reasoning_item: Final = reasoning_done_events[0].item assert getattr(reasoning_item, "status", None) == "completed" - assert getattr(reasoning_item, "encrypted_content", None) == '[{"type":"thinking","thinking":"Let me think","signature":"sig_abc123"}]' + assert ( + getattr(reasoning_item, "encrypted_content", None) + == '[{"type":"thinking","thinking":"Let me think","signature":"sig_abc123"}]' + ) # Verify response.completed carries the exact same encrypted_content and id - completed_event: Final = next(e for e in events if getattr(e, "type", None) == ResponsesAPIStreamEvents.RESPONSE_COMPLETED) + completed_event: Final = next( + e for e in events if getattr(e, "type", None) == ResponsesAPIStreamEvents.RESPONSE_COMPLETED + ) completed_reasoning: Final = next(item for item in completed_event.response.output if item.type == "reasoning") assert completed_reasoning.id == getattr(reasoning_item, "id", None) assert completed_reasoning.status == "completed" @@ -1569,13 +1531,22 @@ async def test_reasoning_output_item_done_encrypted_content_incremental_stream(s encrypted_content on output_item.done. """ chunk1: Final = _chunk("", finish_reason=None) - delta1: Final = Delta(**{**chunk1.choices[0].delta.model_dump(), "thinking_blocks": [{"type": "thinking", "thinking": "Step 1 "}]}) + delta1: Final = Delta( + **{**chunk1.choices[0].delta.model_dump(), "thinking_blocks": [{"type": "thinking", "thinking": "Step 1 "}]} + ) chunk2: Final = _chunk("", finish_reason=None) - delta2: Final = Delta(**{**chunk2.choices[0].delta.model_dump(), "thinking_blocks": [{"type": "thinking", "thinking": "Step 2"}]}) + delta2: Final = Delta( + **{**chunk2.choices[0].delta.model_dump(), "thinking_blocks": [{"type": "thinking", "thinking": "Step 2"}]} + ) chunk3: Final = _chunk("Hello!", finish_reason="stop") - delta3: Final = Delta(**{**chunk3.choices[0].delta.model_dump(), "thinking_blocks": [{"type": "thinking", "thinking": "", "signature": "sig_done"}]}) + delta3: Final = Delta( + **{ + **chunk3.choices[0].delta.model_dump(), + "thinking_blocks": [{"type": "thinking", "thinking": "", "signature": "sig_done"}], + } + ) chunks: Final = ( chunk1.model_copy(update={"choices": [StreamingChoices(index=0, delta=delta1, finish_reason=None)]}), @@ -1594,7 +1565,10 @@ async def test_reasoning_output_item_done_encrypted_content_incremental_stream(s assert len(reasoning_done_events) == 1 reasoning_item: Final = reasoning_done_events[0].item assert getattr(reasoning_item, "status", None) == "completed" - assert getattr(reasoning_item, "encrypted_content", None) == '[{"type":"thinking","thinking":"Step 1 Step 2","signature":"sig_done"}]' + assert ( + getattr(reasoning_item, "encrypted_content", None) + == '[{"type":"thinking","thinking":"Step 1 Step 2","signature":"sig_done"}]' + ) @pytest.mark.parametrize("sync_mode", [True, False]) @@ -1617,4 +1591,3 @@ async def test_reasoning_output_item_done_without_signed_thinking_omits_encrypte reasoning_item: Final = reasoning_done_events[0].item assert getattr(reasoning_item, "status", None) == "completed" assert getattr(reasoning_item, "encrypted_content", None) is None -