feat(proxy_server.py): support maxage cache control

litellm/proxy/proxy_server.py

@@ -995,6 +995,20 @@ def get_litellm_model_info(model: dict = {}):
         return {}
 
 
+def parse_cache_control(cache_control):
+    cache_dict = {}
+    directives = cache_control.split(", ")
+
+    for directive in directives:
+        if "=" in directive:
+            key, value = directive.split("=")
+            cache_dict[key] = value
+        else:
+            cache_dict[directive] = True
+
+    return cache_dict
+
+
 @router.on_event("startup")
 async def startup_event():
     global prisma_client, master_key, use_background_health_checks
@@ -1223,6 +1237,14 @@ async def chat_completion(
             "body": copy.copy(data),  # use copy instead of deepcopy
         }
 
+        ## Cache Controls
+        headers = request.headers
+        print("Request Headers:", headers)
+        cache_control_header = headers.get("Cache-Control", None)
+        if cache_control_header:
+            cache_dict = parse_cache_control(cache_control_header)
+            data["ttl"] = cache_dict.get("s-maxage")
+
         print_verbose(f"receiving data: {data}")
         data["model"] = (
             general_settings.get("completion_model", None)  # server default

litellm/router.py

@@ -1529,7 +1529,6 @@ class Router:
 
             ############ End of initializing Clients for OpenAI/Azure ###################
             self.deployment_names.append(model["litellm_params"]["model"])
-            self.print_verbose(f"\n Initialized Model List {self.model_list}")
 
             ############ Users can either pass tpm/rpm as a litellm_param or a router param ###########
             # for get_available_deployment, we use the litellm_param["rpm"]
@@ -1545,6 +1544,7 @@ class Router:
             ):
                 model["litellm_params"]["tpm"] = model.get("tpm")
 
+        self.print_verbose(f"\nInitialized Model List {self.model_list}")
         self.model_names = [m["model_name"] for m in model_list]
 
     def get_model_names(self):