Litellm remove circular imports (#7232)

* fix(utils.py): initial commit to remove circular imports - moves llmproviders to utils.py

* fix(router.py): fix 'litellm.EmbeddingResponse' import from router.py

'

* refactor: fix litellm.ModelResponse import on pass through endpoints

* refactor(litellm_logging.py): fix circular import for custom callbacks literal

* fix(factory.py): fix circular imports inside prompt factory

* fix(cost_calculator.py): fix circular import for 'litellm.Usage'

* fix(proxy_server.py): fix potential circular import with `litellm.Router'

* fix(proxy/utils.py): fix potential circular import in `litellm.Router`

* fix: remove circular imports in 'auth_checks' and 'guardrails/'

* fix(prompt_injection_detection.py): fix router impor t

* fix(vertex_passthrough_logging_handler.py): fix potential circular imports in vertex pass through

* fix(anthropic_pass_through_logging_handler.py): fix potential circular imports

* fix(slack_alerting.py-+-ollama_chat.py): fix modelresponse import

* fix(base.py): fix potential circular import

* fix(handler.py): fix potential circular ref in codestral + cohere handler's

* fix(azure.py): fix potential circular imports

* fix(gpt_transformation.py): fix modelresponse import

* fix(litellm_logging.py): add logging base class - simplify typing

makes it easy for other files to type check the logging obj without introducing circular imports

* fix(azure_ai/embed): fix potential circular import on handler.py

* fix(databricks/): fix potential circular imports in databricks/

* fix(vertex_ai/): fix potential circular imports on vertex ai embeddings

* fix(vertex_ai/image_gen): fix import

* fix(watsonx-+-bedrock): cleanup imports

* refactor(anthropic-pass-through-+-petals): cleanup imports

* refactor(huggingface/): cleanup imports

* fix(ollama-+-clarifai): cleanup circular imports

* fix(openai_like/): fix impor t

* fix(openai_like/): fix embedding handler

cleanup imports

* refactor(openai.py): cleanup imports

* fix(sagemaker/transformation.py): fix import

* ci(config.yml): add circular import test to ci/cd

.circleci/config.yml

@@ -817,6 +817,7 @@ jobs:
       - run: python ./tests/documentation_tests/test_api_docs.py
       - run: python ./tests/code_coverage_tests/ensure_async_clients_test.py
       - run: python ./tests/code_coverage_tests/enforce_llms_folder_style.py
+      - run: python ./tests/documentation_tests/test_circular_imports.py
       - run: helm lint ./deploy/charts/litellm-helm
 
   db_migration_disable_update_check:

enterprise/enterprise_hooks/secret_detection.py

@@ -474,12 +474,9 @@ async def async_pre_call_hook(
         from detect_secrets import SecretsCollection
         from detect_secrets.settings import default_settings
 
-        print("INSIDE SECRET DETECTION PRE-CALL HOOK!")
-
         if await self.should_run_check(user_api_key_dict) is False:
             return
 
-        print("RUNNING CHECK!")
         if "messages" in data and isinstance(data["messages"], list):
             for message in data["messages"]:
                 if "content" in message and isinstance(message["content"], str):

litellm/__init__.py

@@ -32,7 +32,7 @@
     KeyManagementSettings,
     LiteLLM_UpperboundKeyGenerateParams,
 )
-from litellm.types.utils import StandardKeyGenerationConfig
+from litellm.types.utils import StandardKeyGenerationConfig, LlmProviders
 import httpx
 import dotenv
 from enum import Enum
@@ -838,71 +838,6 @@ def add_known_models():
 )
 
 
-class LlmProviders(str, Enum):
-    OPENAI = "openai"
-    OPENAI_LIKE = "openai_like"  # embedding only
-    JINA_AI = "jina_ai"
-    XAI = "xai"
-    CUSTOM_OPENAI = "custom_openai"
-    TEXT_COMPLETION_OPENAI = "text-completion-openai"
-    COHERE = "cohere"
-    COHERE_CHAT = "cohere_chat"
-    CLARIFAI = "clarifai"
-    ANTHROPIC = "anthropic"
-    ANTHROPIC_TEXT = "anthropic_text"
-    REPLICATE = "replicate"
-    HUGGINGFACE = "huggingface"
-    TOGETHER_AI = "together_ai"
-    OPENROUTER = "openrouter"
-    VERTEX_AI = "vertex_ai"
-    VERTEX_AI_BETA = "vertex_ai_beta"
-    GEMINI = "gemini"
-    AI21 = "ai21"
-    BASETEN = "baseten"
-    AZURE = "azure"
-    AZURE_TEXT = "azure_text"
-    AZURE_AI = "azure_ai"
-    SAGEMAKER = "sagemaker"
-    SAGEMAKER_CHAT = "sagemaker_chat"
-    BEDROCK = "bedrock"
-    VLLM = "vllm"
-    NLP_CLOUD = "nlp_cloud"
-    PETALS = "petals"
-    OOBABOOGA = "oobabooga"
-    OLLAMA = "ollama"
-    OLLAMA_CHAT = "ollama_chat"
-    DEEPINFRA = "deepinfra"
-    PERPLEXITY = "perplexity"
-    MISTRAL = "mistral"
-    GROQ = "groq"
-    NVIDIA_NIM = "nvidia_nim"
-    CEREBRAS = "cerebras"
-    AI21_CHAT = "ai21_chat"
-    VOLCENGINE = "volcengine"
-    CODESTRAL = "codestral"
-    TEXT_COMPLETION_CODESTRAL = "text-completion-codestral"
-    DEEPSEEK = "deepseek"
-    SAMBANOVA = "sambanova"
-    MARITALK = "maritalk"
-    VOYAGE = "voyage"
-    CLOUDFLARE = "cloudflare"
-    XINFERENCE = "xinference"
-    FIREWORKS_AI = "fireworks_ai"
-    FRIENDLIAI = "friendliai"
-    WATSONX = "watsonx"
-    WATSONX_TEXT = "watsonx_text"
-    TRITON = "triton"
-    PREDIBASE = "predibase"
-    DATABRICKS = "databricks"
-    EMPOWER = "empower"
-    GITHUB = "github"
-    CUSTOM = "custom"
-    LITELLM_PROXY = "litellm_proxy"
-    HOSTED_VLLM = "hosted_vllm"
-    LM_STUDIO = "lm_studio"
-    GALADRIEL = "galadriel"
-
-
 provider_list: List[Union[LlmProviders, str]] = list(LlmProviders)
 
 

litellm/adapters/anthropic_adapter.py

@@ -18,7 +18,7 @@
     AnthropicResponse,
     ContentBlockDelta,
 )
-from litellm.types.utils import AdapterCompletionStreamWrapper
+from litellm.types.utils import AdapterCompletionStreamWrapper, ModelResponse
 
 
 class AnthropicAdapter(CustomLogger):
@@ -41,7 +41,7 @@ def translate_completion_input_params(
         return translated_body
 
     def translate_completion_output_params(
-        self, response: litellm.ModelResponse
+        self, response: ModelResponse
     ) -> Optional[AnthropicResponse]:
 
         return litellm.AnthropicExperimentalPassThroughConfig().translate_openai_response_to_anthropic(

litellm/cost_calculator.py

@@ -484,15 +484,15 @@ def completion_cost(  # noqa: PLR0915
         completion_characters: Optional[int] = None
         cache_creation_input_tokens: Optional[int] = None
         cache_read_input_tokens: Optional[int] = None
-        cost_per_token_usage_object: Optional[litellm.Usage] = _get_usage_object(
+        cost_per_token_usage_object: Optional[Usage] = _get_usage_object(
             completion_response=completion_response
         )
         if completion_response is not None and (
             isinstance(completion_response, BaseModel)
             or isinstance(completion_response, dict)
         ):  # tts returns a custom class
 
-            usage_obj: Optional[Union[dict, litellm.Usage]] = completion_response.get(  # type: ignore
+            usage_obj: Optional[Union[dict, Usage]] = completion_response.get(  # type: ignore
                 "usage", {}
             )
             if isinstance(usage_obj, BaseModel) and not isinstance(

litellm/integrations/SlackAlerting/slack_alerting.py

@@ -39,6 +39,7 @@
     VirtualKeyEvent,
     WebhookEvent,
 )
+from litellm.router import Router
 from litellm.types.integrations.slack_alerting import *
 from litellm.types.router import LiteLLM_Params
 
@@ -93,7 +94,7 @@ def update_values(
         alert_types: Optional[List[AlertType]] = None,
         alert_to_webhook_url: Optional[Dict[AlertType, Union[List[str], str]]] = None,
         alerting_args: Optional[Dict] = None,
-        llm_router: Optional[litellm.Router] = None,
+        llm_router: Optional[Router] = None,
     ):
         if alerting is not None:
             self.alerting = alerting

litellm/litellm_core_utils/litellm_logging.py

@@ -18,6 +18,7 @@
 
 import litellm
 from litellm import (
+    _custom_logger_compatible_callbacks_literal,
     json_logs,
     log_raw_request_response,
     turn_off_message_logging,
@@ -41,6 +42,7 @@
     CallTypes,
     EmbeddingResponse,
     ImageResponse,
+    LiteLLMLoggingBaseClass,
     ModelResponse,
     StandardCallbackDynamicParams,
     StandardLoggingAdditionalHeaders,
@@ -190,7 +192,7 @@ def set_cache(self, credentials: dict, service_name: str, logging_obj: Any) -> N
 in_memory_dynamic_logger_cache = DynamicLoggingCache()
 
 
-class Logging:
+class Logging(LiteLLMLoggingBaseClass):
     global supabaseClient, promptLayerLogger, weightsBiasesLogger, logfireLogger, capture_exception, add_breadcrumb, lunaryLogger, logfireLogger, prometheusLogger, slack_app
     custom_pricing: bool = False
     stream_options = None
@@ -2142,7 +2144,7 @@ def set_callbacks(callback_list, function_id=None):  # noqa: PLR0915
 
 
 def _init_custom_logger_compatible_class(  # noqa: PLR0915
-    logging_integration: litellm._custom_logger_compatible_callbacks_literal,
+    logging_integration: _custom_logger_compatible_callbacks_literal,
     internal_usage_cache: Optional[DualCache],
     llm_router: Optional[
         Any
@@ -2362,7 +2364,7 @@ def _init_custom_logger_compatible_class(  # noqa: PLR0915
 
 
 def get_custom_logger_compatible_class(  # noqa: PLR0915
-    logging_integration: litellm._custom_logger_compatible_callbacks_literal,
+    logging_integration: _custom_logger_compatible_callbacks_literal,
 ) -> Optional[CustomLogger]:
     if logging_integration == "lago":
         for callback in _in_memory_loggers:

litellm/litellm_core_utils/prompt_templates/factory.py

@@ -13,7 +13,6 @@
 import litellm
 import litellm.types
 import litellm.types.llms
-import litellm.types.llms.vertex_ai
 from litellm import verbose_logger
 from litellm.llms.custom_httpx.http_handler import HTTPHandler
 from litellm.types.completion import (
@@ -40,6 +39,9 @@
     ChatCompletionUserMessage,
     OpenAIMessageContentListBlock,
 )
+from litellm.types.llms.vertex_ai import FunctionCall as VertexFunctionCall
+from litellm.types.llms.vertex_ai import FunctionResponse as VertexFunctionResponse
+from litellm.types.llms.vertex_ai import PartType as VertexPartType
 from litellm.types.utils import GenericImageParsingChunk
 
 from .common_utils import convert_content_list_to_str, is_non_content_values_set
@@ -965,11 +967,11 @@ def infer_protocol_value(
 
 def _gemini_tool_call_invoke_helper(
     function_call_params: ChatCompletionToolCallFunctionChunk,
-) -> Optional[litellm.types.llms.vertex_ai.FunctionCall]:
+) -> Optional[VertexFunctionCall]:
     name = function_call_params.get("name", "") or ""
     arguments = function_call_params.get("arguments", "")
     arguments_dict = json.loads(arguments)
-    function_call = litellm.types.llms.vertex_ai.FunctionCall(
+    function_call = VertexFunctionCall(
         name=name,
         args=arguments_dict,
     )
@@ -978,7 +980,7 @@ def _gemini_tool_call_invoke_helper(
 
 def convert_to_gemini_tool_call_invoke(
     message: ChatCompletionAssistantMessage,
-) -> List[litellm.types.llms.vertex_ai.PartType]:
+) -> List[VertexPartType]:
     """
     OpenAI tool invokes:
     {
@@ -1019,22 +1021,20 @@ def convert_to_gemini_tool_call_invoke(
     - json.load the arguments
     """
     try:
-        _parts_list: List[litellm.types.llms.vertex_ai.PartType] = []
+        _parts_list: List[VertexPartType] = []
         tool_calls = message.get("tool_calls", None)
         function_call = message.get("function_call", None)
         if tool_calls is not None:
             for tool in tool_calls:
                 if "function" in tool:
-                    gemini_function_call: Optional[
-                        litellm.types.llms.vertex_ai.FunctionCall
-                    ] = _gemini_tool_call_invoke_helper(
-                        function_call_params=tool["function"]
+                    gemini_function_call: Optional[VertexFunctionCall] = (
+                        _gemini_tool_call_invoke_helper(
+                            function_call_params=tool["function"]
+                        )
                     )
                     if gemini_function_call is not None:
                         _parts_list.append(
-                            litellm.types.llms.vertex_ai.PartType(
-                                function_call=gemini_function_call
-                            )
+                            VertexPartType(function_call=gemini_function_call)
                         )
                     else:  # don't silently drop params. Make it clear to user what's happening.
                         raise Exception(
@@ -1047,11 +1047,7 @@ def convert_to_gemini_tool_call_invoke(
                 function_call_params=function_call
             )
             if gemini_function_call is not None:
-                _parts_list.append(
-                    litellm.types.llms.vertex_ai.PartType(
-                        function_call=gemini_function_call
-                    )
-                )
+                _parts_list.append(VertexPartType(function_call=gemini_function_call))
             else:  # don't silently drop params. Make it clear to user what's happening.
                 raise Exception(
                     "function_call missing. Received tool call with 'type': 'function'. No function call in argument - {}".format(
@@ -1070,7 +1066,7 @@ def convert_to_gemini_tool_call_invoke(
 def convert_to_gemini_tool_call_result(
     message: Union[ChatCompletionToolMessage, ChatCompletionFunctionMessage],
     last_message_with_tool_calls: Optional[dict],
-) -> litellm.types.llms.vertex_ai.PartType:
+) -> VertexPartType:
     """
     OpenAI message with a tool result looks like:
     {
@@ -1119,11 +1115,11 @@ def convert_to_gemini_tool_call_result(
 
     # We can't determine from openai message format whether it's a successful or
     # error call result so default to the successful result template
-    _function_response = litellm.types.llms.vertex_ai.FunctionResponse(
+    _function_response = VertexFunctionResponse(
         name=name, response={"content": content_str}  # type: ignore
     )
 
-    _part = litellm.types.llms.vertex_ai.PartType(function_response=_function_response)
+    _part = VertexPartType(function_response=_function_response)
 
     return _part
 

litellm/llms/anthropic/experimental_pass_through/transformation.py

@@ -5,6 +5,12 @@
 from openai.types.chat.chat_completion_chunk import Choice as OpenAIStreamingChoice
 
 import litellm
+from litellm.litellm_core_utils.prompt_templates.factory import (
+    anthropic_messages_pt,
+    custom_prompt,
+    prompt_factory,
+)
+from litellm.litellm_core_utils.streaming_handler import CustomStreamWrapper
 from litellm.types.llms.anthropic import (
     AllAnthropicToolsValues,
     AnthopicMessagesAssistantMessageParam,
@@ -53,15 +59,9 @@
     ChatCompletionUserMessage,
     OpenAIMessageContent,
 )
-from litellm.types.utils import Choices, GenericStreamingChunk
-from litellm.utils import CustomStreamWrapper, ModelResponse, Usage
+from litellm.types.utils import Choices, GenericStreamingChunk, ModelResponse, Usage
 
 from ...base import BaseLLM
-from litellm.litellm_core_utils.prompt_templates.factory import (
-    anthropic_messages_pt,
-    custom_prompt,
-    prompt_factory,
-)
 
 
 class AnthropicExperimentalPassThroughConfig:
@@ -338,7 +338,7 @@ def _translate_openai_finish_reason_to_anthropic(
         return "end_turn"
 
     def translate_openai_response_to_anthropic(
-        self, response: litellm.ModelResponse
+        self, response: ModelResponse
     ) -> AnthropicResponse:
         ## translate content block
         anthropic_content = self._translate_openai_content_to_anthropic(choices=response.choices)  # type: ignore
@@ -347,7 +347,7 @@ def translate_openai_response_to_anthropic(
             openai_finish_reason=response.choices[0].finish_reason  # type: ignore
         )
         # extract usage
-        usage: litellm.Usage = getattr(response, "usage")
+        usage: Usage = getattr(response, "usage")
         anthropic_usage = AnthropicResponseUsageBlock(
             input_tokens=usage.prompt_tokens or 0,
             output_tokens=usage.completion_tokens or 0,
@@ -393,7 +393,7 @@ def _translate_streaming_openai_chunk_to_anthropic(
             return "text_delta", ContentTextBlockDelta(type="text_delta", text=text)
 
     def translate_streaming_openai_response_to_anthropic(
-        self, response: litellm.ModelResponse
+        self, response: ModelResponse
     ) -> Union[ContentBlockDelta, MessageBlockDelta]:
         ## base case - final chunk w/ finish reason
         if response.choices[0].finish_reason is not None:
@@ -403,7 +403,7 @@ def translate_streaming_openai_response_to_anthropic(
                 ),
             )
             if getattr(response, "usage", None) is not None:
-                litellm_usage_chunk: Optional[litellm.Usage] = response.usage  # type: ignore
+                litellm_usage_chunk: Optional[Usage] = response.usage  # type: ignore
             elif (
                 hasattr(response, "_hidden_params")
                 and "usage" in response._hidden_params