Merge pull request #5029 from BerriAI/litellm_azure_ui_fix

fix(utils.py): Fix adding azure models on ui

litellm/proxy/_new_secret_config.yaml

@@ -1,8 +1,7 @@
 model_list:
-  - model_name: "predibase-llama"
+  - model_name: "*"
     litellm_params:
-      model: "predibase/llama-3-8b-instruct"
-      request_timeout: 1
+      model: "*"
 
-litellm_settings:
-  failure_callback: ["langfuse"]
+# litellm_settings:
+#   failure_callback: ["langfuse"]

litellm/proxy/hooks/parallel_request_limiter.py

@@ -472,11 +472,10 @@ class _PROXY_MaxParallelRequestsHandler(CustomLogger):
 
     async def async_log_failure_event(self, kwargs, response_obj, start_time, end_time):
         try:
-            self.print_verbose(f"Inside Max Parallel Request Failure Hook")
-            global_max_parallel_requests = (
-                kwargs["litellm_params"]
-                .get("metadata", {})
-                .get("global_max_parallel_requests", None)
+            self.print_verbose("Inside Max Parallel Request Failure Hook")
+            _metadata = kwargs["litellm_params"].get("metadata", {}) or {}
+            global_max_parallel_requests = _metadata.get(
+                "global_max_parallel_requests", None
             )
             user_api_key = (
                 kwargs["litellm_params"].get("metadata", {}).get("user_api_key", None)

litellm/proxy/proxy_server.py

@@ -1959,6 +1959,7 @@ class ProxyConfig:
                         if len(_value) > 0:
                             _litellm_params[k] = _value
                 _litellm_params = LiteLLM_Params(**_litellm_params)
+
             else:
                 verbose_proxy_logger.error(
                     f"Invalid model added to proxy db. Invalid litellm params. litellm_params={_litellm_params}"

litellm/types/utils.py

@@ -304,7 +304,7 @@ class Message(OpenAIObject):
         content: Optional[str] = None,
         role: Literal["assistant"] = "assistant",
         function_call=None,
-        tool_calls=None,
+        tool_calls: Optional[list] = None,
         **params,
     ):
         init_values = {
@@ -322,7 +322,7 @@ class Message(OpenAIObject):
                     )
                     for tool_call in tool_calls
                 ]
-                if tool_calls is not None
+                if tool_calls is not None and len(tool_calls) > 0
                 else None
             ),
         }
@@ -445,8 +445,6 @@ class Choices(OpenAIObject):
 
 
 class Usage(OpenAIObject):
-    prompt_cache_hit_tokens: Optional[int] = Field(default=None)
-    prompt_cache_miss_tokens: Optional[int] = Field(default=None)
     prompt_tokens: Optional[int] = Field(default=None)
     completion_tokens: Optional[int] = Field(default=None)
     total_tokens: Optional[int] = Field(default=None)
@@ -456,16 +454,15 @@ class Usage(OpenAIObject):
         prompt_tokens: Optional[int] = None,
         completion_tokens: Optional[int] = None,
         total_tokens: Optional[int] = None,
-        prompt_cache_hit_tokens: Optional[int] = None,
-        prompt_cache_miss_tokens: Optional[int] = None,
+        **params,
     ):
         data = {
             "prompt_tokens": prompt_tokens,
             "completion_tokens": completion_tokens,
             "total_tokens": total_tokens,
-            "prompt_cache_hit_tokens": prompt_cache_hit_tokens,
-            "prompt_cache_miss_tokens": prompt_cache_miss_tokens,
+            **params,
         }
+
         super().__init__(**data)
 
     def __contains__(self, key):

litellm/utils.py

@@ -4446,6 +4446,11 @@ def get_llm_provider(
                 return model, custom_llm_provider, dynamic_api_key, api_base
 
         if custom_llm_provider:
+            if (
+                model.split("/")[0] == custom_llm_provider
+            ):  # handle scenario where model="azure/*" and custom_llm_provider="azure"
+                model = model.replace("{}/".format(custom_llm_provider), "")
+
             return model, custom_llm_provider, dynamic_api_key, api_base
 
         if api_key and api_key.startswith("os.environ/"):
@@ -5827,9 +5832,10 @@ def convert_to_model_response_object(
                 model_response_object.usage.completion_tokens = response_object["usage"].get("completion_tokens", 0)  # type: ignore
                 model_response_object.usage.prompt_tokens = response_object["usage"].get("prompt_tokens", 0)  # type: ignore
                 model_response_object.usage.total_tokens = response_object["usage"].get("total_tokens", 0)  # type: ignore
-                model_response_object.usage.prompt_cache_hit_tokens = response_object["usage"].get("prompt_cache_hit_tokens", None)  # type: ignore
-                model_response_object.usage.prompt_cache_miss_tokens = response_object["usage"].get("prompt_cache_miss_tokens", None)  # type: ignore
-
+                special_keys = ["completion_tokens", "prompt_tokens", "total_tokens"]
+                for k, v in response_object["usage"].items():
+                    if k not in special_keys:
+                        setattr(model_response_object.usage, k, v)  # type: ignore
             if "created" in response_object:
                 model_response_object.created = response_object["created"] or int(
                     time.time()