From 6770767d8a6098c33873f9ef815bb0d6b7c1e3f8 Mon Sep 17 00:00:00 2001
From: Concedo <39025047+LostRuins@users.noreply.github.com>
Date: Thu, 27 Nov 2025 01:03:29 +0800
Subject: [PATCH] allow FA for clip but with wmma disabled for turing on bad
 sizes

---
 gpttype_adapter.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/gpttype_adapter.cpp b/gpttype_adapter.cpp
index aacc29222..9df8a4178 100644
--- a/gpttype_adapter.cpp
+++ b/gpttype_adapter.cpp
@@ -2470,7 +2470,7 @@ ModelLoadResult gpttype_load_model(const load_model_inputs inputs, FileFormat in
             }
             clip_context_params ctx_clip_params {
                 /* use_gpu           */ true,
-                /* flash_attn_type   */ CLIP_FLASH_ATTN_TYPE_DISABLED,
+                /* flash_attn_type   */ (kcpp_data->flash_attn?CLIP_FLASH_ATTN_TYPE_ENABLED:CLIP_FLASH_ATTN_TYPE_DISABLED),
                 /* image_min_tokens  */ -1,
                 /* image_max_tokens  */ -1,
             };