From a884bc62d6eeeb58e87e56b2d966c63f073ada77 Mon Sep 17 00:00:00 2001
From: Xin Yang <105740670+xyang16@users.noreply.github.com>
Date: Fri, 16 Jan 2026 10:38:07 -0800
Subject: [PATCH] [LoRA] Update LoRA expand kernel heuristic (#32425)

Signed-off-by: Xin Yang <xyangx@amazon.com>
---
 vllm/lora/ops/triton_ops/utils.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/vllm/lora/ops/triton_ops/utils.py b/vllm/lora/ops/triton_ops/utils.py
index 51535f32c1..f0cb9a5c04 100644
--- a/vllm/lora/ops/triton_ops/utils.py
+++ b/vllm/lora/ops/triton_ops/utils.py
@@ -251,7 +251,7 @@ def get_lora_op_configs(
     else:
         default = {
             "block_m": 64,
-            "block_n": 128,
+            "block_n": max(64, next_power_of_2(128 // num_slices)),
             "block_k": 16,
             "num_warps": 4,
             "num_ctas": 1,