From d68f4a5ae51ec132e81aa0f3ca8417c7144254b1 Mon Sep 17 00:00:00 2001
From: Concedo <39025047+LostRuins@users.noreply.github.com>
Date: Thu, 27 Nov 2025 10:20:38 +0800
Subject: [PATCH] disable clip fa for now

---
 gpttype_adapter.cpp | 2 +-
 koboldcpp.py        | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/gpttype_adapter.cpp b/gpttype_adapter.cpp
index 9df8a4178..70f4be43e 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   */ (kcpp_data->flash_attn?CLIP_FLASH_ATTN_TYPE_ENABLED:CLIP_FLASH_ATTN_TYPE_DISABLED),
+                /* flash_attn_type   */ CLIP_FLASH_ATTN_TYPE_DISABLED, //kcpp: disabled in 1.102.2 as some headsizes break on turing
                 /* image_min_tokens  */ -1,
                 /* image_max_tokens  */ -1,
             };
diff --git a/koboldcpp.py b/koboldcpp.py
index 09834fafe..ba1fc5093 100755
--- a/koboldcpp.py
+++ b/koboldcpp.py
@@ -66,7 +66,7 @@ dry_seq_break_max = 128
 extra_images_max = 4 # for kontext/qwen img
 
 # global vars
-KcppVersion = "1.102.1"
+KcppVersion = "1.102.2"
 showdebug = True
 kcpp_instance = None #global running instance
 global_memory = {"tunnel_url": "", "restart_target":"", "input_to_exit":False, "load_complete":False, "restart_override_config_target":""}