Set `max_train_steps` to 0 if not specified in older `.json` config files

.release

@@ -1 +1 @@
-v24.0.7
+v24.0.8

README.md

@@ -47,6 +47,7 @@ The GUI allows you to set the training parameters and generate and run the requi
   - [SDXL training](#sdxl-training)
   - [Masked loss](#masked-loss)
   - [Change History](#change-history)
+    - [2024/04/26 (v24.0.8)](#20240426-v2408)
     - [2024/04/25 (v24.0.7)](#20240425-v2407)
     - [2024/04/22 (v24.0.6)](#20240422-v2406)
     - [2024/04/19 (v24.0.5)](#20240419-v2405)
@@ -453,6 +454,10 @@ ControlNet dataset is used to specify the mask. The mask images should be the RG
 
 ## Change History
 
+### 2024/04/26 (v24.0.8)
+
+- Set `max_train_steps` to 0 if not specified in older `.json` config files.
+
 ### 2024/04/25 (v24.0.7)
 
 - Prevent crash if tkinter is not installed

kohya_gui/common_gui.py

@@ -364,7 +364,7 @@ def update_my_data(my_data):
                 my_data[key] = int(value)
             except ValueError:
                 # Handle the case where the string is not a valid float
-                my_data[key] = int(1600)
+                my_data[key] = int(0)
 
     # Convert values to int if they are strings
     for key in ["max_token_length"]:

test/config/dreambooth-AdamW8bit.json

@@ -53,7 +53,7 @@
   "max_timestep": 1000,
   "max_token_length": 75,
   "max_train_epochs": 0,
-  "max_train_steps": 0,
+  "max_train_steps": "",
   "mem_eff_attn": false,
   "min_bucket_reso": 256,
   "min_snr_gamma": 0,