From b7428048fce19c24fcaa387c74ace4f20eac4e05 Mon Sep 17 00:00:00 2001
From: Concedo <39025047+LostRuins@users.noreply.github.com>
Date: Thu, 11 Dec 2025 14:30:38 +0800
Subject: [PATCH] try reduce pipeline parallelism in order to reduce compute
 buffer sizes

---
 ggml/src/ggml-backend.cpp | 3 ++-
 koboldcpp.py              | 2 +-
 2 files changed, 3 insertions(+), 2 deletions(-)

diff --git a/ggml/src/ggml-backend.cpp b/ggml/src/ggml-backend.cpp
index 8b90a7d1e..84041d00f 100644
--- a/ggml/src/ggml-backend.cpp
+++ b/ggml/src/ggml-backend.cpp
@@ -662,7 +662,8 @@ static bool ggml_is_view_op(enum ggml_op op) {
 #endif
 
 #ifndef GGML_SCHED_MAX_COPIES
-#define GGML_SCHED_MAX_COPIES 4
+//kcpp reduced from 4 to 2 to try make buffer sizes smaller on multigpu
+#define GGML_SCHED_MAX_COPIES 2
 #endif
 
 struct ggml_backend_sched_split {
diff --git a/koboldcpp.py b/koboldcpp.py
index 1c1ce2eed..9d407a801 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 = "1.103"
+KcppVersion = "1.104"
 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":""}