Merge pull request #2398 from bmaltais/dev

v24.0.8

.env

@@ -0,0 +1 @@
+TENSORBOARD_PORT=6006

.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)
@@ -105,7 +106,7 @@ To set up the project, follow these steps:
 2. Clone the repository by running the following command:
 
    ```shell
-   git clone https://github.com/bmaltais/kohya_ss.git
+   git clone --recursive https://github.com/bmaltais/kohya_ss.git
    ```
 
 3. Change into the `kohya_ss` directory:
@@ -153,7 +154,7 @@ To set up the project on Linux or macOS, perform the following steps:
 2. Clone the repository by running the following command:
 
    ```shell
-   git clone https://github.com/bmaltais/kohya_ss.git
+   git clone --recursive https://github.com/bmaltais/kohya_ss.git
    ```
 
 3. Change into the `kohya_ss` directory:
@@ -198,7 +199,7 @@ To install the necessary components for Runpod and run kohya_ss, follow these st
 
    ```shell
    cd /workspace
-   git clone https://github.com/bmaltais/kohya_ss.git
+   git clone --recursive https://github.com/bmaltais/kohya_ss.git
    ```
 
 4. Run the setup script:
@@ -266,7 +267,7 @@ Install the NVIDIA Container Toolkit with this guide.
 #### Use the pre-built Docker image
 
 ```bash
-git clone https://github.com/bmaltais/kohya_ss.git
+git clone --recursive https://github.com/bmaltais/kohya_ss.git
 cd kohya_ss
 docker compose up -d
 ```
@@ -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

docker-compose.yaml

@@ -15,6 +15,7 @@ services:
       - 7860:7860
     environment:
       SAFETENSORS_FAST_GPU: 1
+      TENSORBOARD_PORT: ${TENSORBOARD_PORT:-6006}
     tmpfs:
       - /tmp
     volumes:
@@ -42,7 +43,8 @@ services:
     container_name: tensorboard
     image: tensorflow/tensorflow:latest-gpu
     ports:
-      - 6006:6006
+      # !Please change the port in .env file
+      - ${TENSORBOARD_PORT:-6006}:6006
     volumes:
       - ./dataset/logs:/app/logs
     command: tensorboard --logdir=/app/logs --bind_all

kohya_gui/class_tensorboard.py

@@ -39,6 +39,11 @@ def get_button_states(self, started=False):
             visible=visibility and (not started or self.headless)
         ), gr.Button(visible=visibility and (started or self.headless))
 
+    def open_tensorboard_url(self):
+        tensorboard_url = f"http://localhost:{self.tensorboard_port}"
+        self.log.info(f"Opening TensorBoard URL in browser: {tensorboard_url}")
+        webbrowser.open(tensorboard_url)
+
     def start_tensorboard(self, logging_dir=None):
         if self.tensorboard_proc is not None:
             self.log.info(
@@ -74,17 +79,13 @@ def start_tensorboard(self, logging_dir=None):
             self.log.error("Failed to start Tensorboard:", e)
             return self.get_button_states(started=False)
 
-        def open_tensorboard_url():
-            time.sleep(self.wait_time)
-            if not self.stop_event.is_set():
-                tensorboard_url = f"http://localhost:{self.tensorboard_port}"
-                self.log.info(f"Opening TensorBoard URL in browser: {tensorboard_url}")
-                webbrowser.open(tensorboard_url)
-
         if not self.headless:
             self.stop_event.clear()
-            self.thread = Thread(target=open_tensorboard_url)
-            self.thread.start()
+
+            time.sleep(self.wait_time)
+            if not self.stop_event.is_set():
+                self.thread = Thread(target=self.open_tensorboard_url)
+                self.thread.start()
 
         return self.get_button_states(started=True)
 
@@ -119,6 +120,12 @@ def gradio_interface(self):
                 visible=visibility and self.headless,
                 elem_id="myTensorButtonStop",
             )
+            button_open_tensorboard = gr.Button(
+                value="Open tensorboard",
+                elem_id="myTensorButton",
+                visible=not visibility,
+                link=f"http://localhost:{self.tensorboard_port}",
+            )
             button_start_tensorboard.click(
                 self.start_tensorboard,
                 inputs=[self.logging_dir],

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,