From b5d9a86e4cc3136234614ef919a4e822a9060c36 Mon Sep 17 00:00:00 2001 From: ishandhanani <82981111+ishandhanani@users.noreply.github.com> Date: Mon, 20 Apr 2026 12:04:35 -0500 Subject: [PATCH] fix: add back priorty as radix cache policy (#23275) --- python/sglang/srt/server_args.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/python/sglang/srt/server_args.py b/python/sglang/srt/server_args.py index db35346a1..1b62823e3 100644 --- a/python/sglang/srt/server_args.py +++ b/python/sglang/srt/server_args.py @@ -201,7 +201,7 @@ FP4_GEMM_RUNNER_BACKEND_CHOICES = [ "flashinfer_trtllm", ] -RADIX_EVICTION_POLICY_CHOICES = ["lru", "lfu", "slru"] +RADIX_EVICTION_POLICY_CHOICES = ["lru", "lfu", "slru", "priority"] RL_ON_POLICY_TARGET_CHOICES = ["fsdp"] @@ -4391,7 +4391,7 @@ class ServerArgs: type=str, choices=RADIX_EVICTION_POLICY_CHOICES, default=ServerArgs.radix_eviction_policy, - help="The eviction policy of radix trees. 'lru' stands for Least Recently Used, 'lfu' stands for Least Frequently Used, and 'slru' stands for Segmented Least Recently Used.", + help="The eviction policy of radix trees. 'lru' stands for Least Recently Used, 'lfu' stands for Least Frequently Used, 'slru' stands for Segmented Least Recently Used, and 'priority' evicts lower-priority requests first.", ) parser.add_argument( "--enable-prefill-delayer",