.

Author: edwin-onyx

backend/onyx/llm/chat_llm.py

@@ -175,17 +175,15 @@ def _convert_delta_to_message_chunk(
     stop_reason: str | None = None,
 ) -> BaseMessageChunk:
     configure_litellm()
-    import litellm
+    from litellm.utils import ChatCompletionDeltaToolCall
 
     """Adapted from langchain_community.chat_models.litellm._convert_delta_to_message_chunk"""
     role = _dict.get("role") or (_base_msg_to_role(curr_msg) if curr_msg else "unknown")
     content = _dict.get("content") or ""
     additional_kwargs = {}
     if _dict.get("function_call"):
         additional_kwargs.update({"function_call": dict(_dict["function_call"])})
-    tool_calls = cast(
-        list[litellm.utils.ChatCompletionDeltaToolCall] | None, _dict.get("tool_calls")
-    )
+    tool_calls = cast(list[ChatCompletionDeltaToolCall] | None, _dict.get("tool_calls"))
 
     if role == "user":
         return HumanMessageChunk(content=content)
@@ -323,9 +321,7 @@ def __init__(
         self._max_token_param = LEGACY_MAX_TOKENS_KWARG
         try:
             configure_litellm()
-            import litellm
-
-            get_supported_openai_params = litellm.utils.get_supported_openai_params
+            from litellm.utils import get_supported_openai_params
 
             params = get_supported_openai_params(model_name, model_provider)
             if STANDARD_MAX_TOKENS_KWARG in (params or []):

backend/scripts/check_lazy_imports.py

@@ -19,9 +19,9 @@
 
 @dataclass
 class LazyImportSettings:
-    """Settings for which directories to ignore when checking for lazy imports."""
+    """Settings for which files to ignore when checking for lazy imports."""
 
-    ignore_directories: Set[str] | None = None
+    ignore_files: Set[str] | None = None
 
 
 # Map of modules to lazy import -> settings for what to ignore
@@ -31,9 +31,7 @@ class LazyImportSettings:
     "markitdown": LazyImportSettings(),
     "tiktoken": LazyImportSettings(),
     "unstructured": LazyImportSettings(),
-    "litellm": LazyImportSettings(
-        ignore_directories={"onyx/llm/llm_provider_options.py"}
-    ),
+    "litellm": LazyImportSettings(ignore_files={"onyx/llm/llm_provider_options.py"}),
 }
 
 
@@ -150,12 +148,12 @@ def should_check_file_for_module(
     Args:
         file_path: Path to the file to check
         backend_dir: Path to the backend directory
-        settings: Settings containing directories to ignore for this module
+        settings: Settings containing files to ignore for this module
 
     Returns:
         True if the file should be checked, False if it should be ignored
     """
-    if not settings.ignore_directories:
+    if not settings.ignore_files:
         # Empty set means check everywhere
         return True
 
@@ -164,7 +162,7 @@ def should_check_file_for_module(
     rel_path_str = str(rel_path)
 
     # Check if this specific file path is in the ignore list
-    return rel_path_str not in settings.ignore_directories
+    return rel_path_str not in settings.ignore_files
 
 
 def main(modules_to_lazy_import: Dict[str, LazyImportSettings]) -> None:

backend/tests/unit/onyx/llm/test_chat_llm.py

@@ -45,7 +45,7 @@ def default_multi_llm() -> DefaultMultiLLM:
 
 def test_multiple_tool_calls(default_multi_llm: DefaultMultiLLM) -> None:
     # Mock the litellm.completion function
-    with patch("onyx.llm.chat_llm.litellm.completion") as mock_completion:
+    with patch("litellm.completion") as mock_completion:
         # Create a mock response with multiple tool calls using litellm objects
         mock_response = litellm.ModelResponse(
             id="chatcmpl-123",
@@ -158,7 +158,7 @@ def test_multiple_tool_calls(default_multi_llm: DefaultMultiLLM) -> None:
 
 def test_multiple_tool_calls_streaming(default_multi_llm: DefaultMultiLLM) -> None:
     # Mock the litellm.completion function
-    with patch("onyx.llm.chat_llm.litellm.completion") as mock_completion:
+    with patch("litellm.completion") as mock_completion:
         # Create a mock response with multiple tool calls using litellm objects
         mock_response = [
             litellm.ModelResponse(