From a682ac09aee2f913cebf3a564361a540501b08c2 Mon Sep 17 00:00:00 2001 From: 72004 Date: Fri, 21 Aug 2026 18:46:35 +0500 Subject: [PATCH] fix(router): preserve deployment retry budget with policy --- litellm/router.py | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/litellm/router.py b/litellm/router.py index e9eeab53934..f6b982f3f62 100644 --- a/litellm/router.py +++ b/litellm/router.py @@ -6703,7 +6703,10 @@ class Router: # Check retry policy FIRST, before should_retry_this_error # This allows retry policies to override the healthy deployments check _retry_policy_applies = False - if self.retry_policy is not None or model_group_retry_policy is not None: + if ( + deployment_num_retries is None + and (self.retry_policy is not None or model_group_retry_policy is not None) + ): # get num_retries from retry policy # Use the model_group captured at the start of the function, or get it from metadata # kwargs.get("model") at this point is the deployment model, not the model_group