From aa5207f5c62aa7c283a1d9d3cb380956c4e0974b Mon Sep 17 00:00:00 2001
From: ebolam <ebolam@gmail.com>
Date: Wed, 30 Nov 2022 07:21:16 -0500
Subject: [PATCH] Ensure we don't save 8bit models

---
 aiserver.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/aiserver.py b/aiserver.py
index 303977e8..f5f7c048 100644
--- a/aiserver.py
+++ b/aiserver.py
@@ -2962,7 +2962,7 @@ def load_model(use_gpu=True, gpu_layers=None, disk_layers=None, initial_load=Fal
 
                         torch._utils._rebuild_tensor = old_rebuild_tensor
 
-                        if not args.colab or args.savemodel:
+                        if (not args.colab or args.savemodel) and not use_8_bit:
                             import shutil
                             tokenizer.save_pretrained("models/{}".format(koboldai_vars.model.replace('/', '_')))
                             if koboldai_vars.fp32_model:  # Use save_pretrained to convert fp32 models to fp16