fix(proxy_cli.py): don't double load the router config

was causing callbacks to be instantiated twice - double couting usage in cache

litellm/proxy/_new_secret_config.yaml

@@ -26,13 +26,12 @@ litellm_settings:
   success_callback: ["prometheus"]
   upperbound_key_generate_params: 
     max_budget: os.environ/LITELLM_UPPERBOUND_KEYS_MAX_BUDGET
-  
-# litellm_settings:
-#   drop_params: True
-#   max_budget: 800021
-#   budget_duration: 30d
-#   # cache: true
-  
+
+router_settings:
+  routing_strategy: usage-based-routing
+  redis_host: redis-16337.c322.us-east-1-2.ec2.cloud.redislabs.com
+  redis_password: madeBerri@992
+  redis_port: 16337
 
 general_settings:
   master_key: sk-1234

litellm/proxy/proxy_cli.py

@@ -425,9 +425,10 @@ def run_server(
                 )
 
             proxy_config = ProxyConfig()
-            _, _, general_settings = asyncio.run(
-                proxy_config.load_config(router=None, config_file_path=config)
-            )
+            _config = asyncio.run(proxy_config.get_config(config_file_path=config))
+            general_settings = _config.get("general_settings", {})
+            if general_settings is None:
+                general_settings = {}
             database_url = general_settings.get("database_url", None)
             db_connection_pool_limit = general_settings.get(
                 "database_connection_pool_limit", 100

litellm/proxy/proxy_server.py

@@ -2335,6 +2335,7 @@ class ProxyConfig:
                 "background_health_checks", False
             )
             health_check_interval = general_settings.get("health_check_interval", 300)
+
         router_params: dict = {
             "cache_responses": litellm.cache
             != None,  # cache if user passed in cache values

litellm/router.py

@@ -2374,7 +2374,6 @@ class Router:
         """
         Returns the deployment based on routing strategy
         """
-
         # users need to explicitly call a specific deployment, by setting `specific_deployment = True` as completion()/embedding() kwarg
         # When this was no explicit we had several issues with fallbacks timing out
         if specific_deployment == True:

litellm/utils.py

@@ -236,7 +236,7 @@ class HiddenParams(OpenAIObject):
 
     class Config:
         extra = "allow"
-        protected_namespaces = ()      
+        protected_namespaces = ()
 
     def get(self, key, default=None):
         # Custom .get() method to access attributes with a default value if the attribute doesn't exist
@@ -1990,9 +1990,6 @@ class Logging:
                             else:
                                 litellm.cache.add_cache(result, **kwargs)
                 if isinstance(callback, CustomLogger):  # custom logger class
-                    print_verbose(
-                        f"Running Async success callback: {callback}; self.stream: {self.stream}; async_complete_streaming_response: {self.model_call_details.get('async_complete_streaming_response', None)} result={result}"
-                    )
                     if self.stream == True:
                         if (
                             "async_complete_streaming_response"
@@ -2376,7 +2373,6 @@ def client(original_function):
             if litellm.use_client or (
                 "use_client" in kwargs and kwargs["use_client"] == True
             ):
-                print_verbose(f"litedebugger initialized")
                 if "lite_debugger" not in litellm.input_callback:
                     litellm.input_callback.append("lite_debugger")
                 if "lite_debugger" not in litellm.success_callback: