From 7c5d47f68882c5c3033b47bb002ad506c2268055 Mon Sep 17 00:00:00 2001
From: Concedo <39025047+LostRuins@users.noreply.github.com>
Date: Thu, 8 May 2025 00:55:09 +0800
Subject: [PATCH] multigpu warning only once

---
 koboldcpp.py | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/koboldcpp.py b/koboldcpp.py
index 40e9bbe39..2a99d2d89 100644
--- a/koboldcpp.py
+++ b/koboldcpp.py
@@ -78,6 +78,7 @@ defaultport = 5001
 showsamplerwarning = True
 showmaxctxwarning = True
 showusedmemwarning = True
+showmultigpuwarning = True
 session_kudos_earned = 0
 session_jobs = 0
 session_starttime = None
@@ -950,7 +951,7 @@ def extract_modelfile_params(filepath,sdfilepath,whisperfilepath,mmprojfilepath,
             modelfile_extracted_meta = None
 
 def autoset_gpu_layers(ctxsize, sdquanted, bbs, qkv_level): #shitty algo to determine how many layers to use
-    global showusedmemwarning, modelfile_extracted_meta # reference cached values instead
+    global showusedmemwarning, showmultigpuwarning, modelfile_extracted_meta # reference cached values instead
     gpumem = MaxMemory[0]
     usedmem = 0
     if MaxFreeMemory[0]>0:
@@ -973,7 +974,9 @@ def autoset_gpu_layers(ctxsize, sdquanted, bbs, qkv_level): #shitty algo to dete
                 if match:
                     total_parts = int(match.group(2))
                     if total_parts > 1 and total_parts <= 9:
-                        print("Multi-Part GGUF detected. Layer estimates may not be very accurate - recommend setting layers manually.")
+                        if showmultigpuwarning:
+                            showmultigpuwarning = False
+                            print("Multi-Part GGUF detected. Layer estimates may not be very accurate - recommend setting layers manually.")
                         fsize *= total_parts
             if modelfile_extracted_meta[3] > 1024*1024*1024*5: #sdxl tax
                 mem -= 1024*1024*1024*(6 if sdquanted else 9)