From 3c6764efef2e2dff5610b3fbf00d49224f4e380c Mon Sep 17 00:00:00 2001 From: ishaan-jaff Date: Wed, 29 Nov 2023 16:22:04 -0800 Subject: [PATCH] (feat) proxy+ router: support 1k request/second --- litellm/router.py | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/litellm/router.py b/litellm/router.py index 5e0605da59..ac0058713e 100644 --- a/litellm/router.py +++ b/litellm/router.py @@ -106,6 +106,14 @@ class Router: ### HEALTH CHECK THREAD ### if self.routing_strategy == "least-busy": self._start_health_check_thread() + if self.routing_strategy == "simple-shuffle": + # use rpm based shuffle if user provided values of rpm + try: + rpm = self.model_list[0]["litellm_params"].get("rpm", None) + if rpm is not None: + self.routing_strategy = "weighted-shuffle" + except: + pass ### CACHING ### redis_cache = None @@ -135,6 +143,7 @@ class Router: litellm.failure_callback.append(self.deployment_callback_on_failure) else: litellm.failure_callback = [self.deployment_callback_on_failure] + self.print_verbose(f"Intialized router with Routing strategy: {self.routing_strategy}\n") ### COMPLETION + EMBEDDING FUNCTIONS