Add parameters to be able to set then on colab demo

TTS/demos/xtts_ft_demo/utils/gpt_train.py

@@ -8,9 +8,9 @@ from TTS.tts.layers.xtts.trainer.gpt_trainer import GPTArgs, GPTTrainer, GPTTrai
 from TTS.utils.manage import ModelManager
 
 
-def train_gpt(language, num_epochs, batch_size, train_csv, eval_csv, output_path):
+def train_gpt(language, num_epochs, batch_size, grad_acumm, train_csv, eval_csv, output_path):
     #  Logging parameters
-    RUN_NAME = "GPT_XTTSv2.1_FT"
+    RUN_NAME = "GPT_XTTS_FT"
     PROJECT_NAME = "XTTS_trainer"
     DASHBOARD_LOGGER = "tensorboard"
     LOGGER_URI = None
@@ -18,13 +18,11 @@ def train_gpt(language, num_epochs, batch_size, train_csv, eval_csv, output_path
     # Set here the path that the checkpoints will be saved. Default: ./run/training/
     OUT_PATH = os.path.join(output_path, "run", "training")
 
-
     # Training Parameters
     OPTIMIZER_WD_ONLY_ON_WEIGHTS = True  # for multi-gpu training please make it False
-    START_WITH_EVAL = True  # if True it will star with evaluation
+    START_WITH_EVAL = False  # if True it will star with evaluation
     BATCH_SIZE = batch_size  # set here the batch size
-    GRAD_ACUMM_STEPS = 1  # set here the grad accumulation steps
-    # Note: we recommend that BATCH_SIZE * GRAD_ACUMM_STEPS need to be at least 252 for more efficient training. You can increase/decrease BATCH_SIZE but then set GRAD_ACUMM_STEPS accordingly.
+    GRAD_ACUMM_STEPS = grad_acumm  # set here the grad accumulation steps
 
 
     # Define here the dataset that you want to use for the fine-tuning on.

TTS/demos/xtts_ft_demo/xtts_demo.py

@@ -1,3 +1,4 @@
+import argparse
 import os
 import sys
 import tempfile
@@ -21,7 +22,6 @@ def clear_gpu_cache():
     if torch.cuda.is_available():
         torch.cuda.empty_cache()
 
-PORT = 5003
 
 XTTS_MODEL = None
 def load_model(xtts_checkpoint, xtts_config, xtts_vocab):
@@ -101,11 +101,54 @@ def read_logs():
         return f.read()
 
 
-with gr.Blocks() as demo:
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser(
+        description="""XTTS fine-tuning demo\n\n"""
+        """
+        Example runs:
+        python3 TTS/demos/xtts_ft_demo/xtts_demo.py --port 
+        """,
+        formatter_class=argparse.RawTextHelpFormatter,
+    )
+    parser.add_argument(
+        "--port",
+        type=int,
+        help="Port to run the gradio demo. Default: 5003",
+        default=5003,
+    )
+    parser.add_argument(
+        "--out_path",
+        type=str,
+        help="Output path (where data and checkpoints will be saved) Default: /tmp/xtts_ft/",
+        default="/tmp/xtts_ft/",
+    )
+
+    parser.add_argument(
+        "--num_epochs",
+        type=int,
+        help="Number of epochs to train. Default: 10",
+        default=10,
+    )
+    parser.add_argument(
+        "--batch_size",
+        type=int,
+        help="Batch size. Default: 4",
+        default=4,
+    )
+    parser.add_argument(
+        "--grad_acumm",
+        type=int,
+        help="Grad accumulation steps. Default: 1",
+        default=1,
+    )
+    args = parser.parse_args()
+
+    with gr.Blocks() as demo:
         with gr.Tab("Data processing"):
             out_path = gr.Textbox(
                 label="Output path (where data and checkpoints will be saved):",
-            value="/tmp/xtts_ft/"
+                value=args.out_path,
             )
             # upload_file = gr.Audio(
             #     sources="upload",
@@ -178,18 +221,25 @@ with gr.Blocks() as demo:
                 label="Eval CSV:",
             )
             num_epochs =  gr.Slider(
-            label="num_epochs",
+                label="Number of epochs:",
                 minimum=1,
                 maximum=100,
                 step=1,
-            value=10,
+                value=args.num_epochs,
             )
             batch_size = gr.Slider(
-            label="batch_size",
+                label="Batch size:",
                 minimum=2,
                 maximum=512,
                 step=1,
-            value=4,
+                value=args.batch_size,
+            )
+            grad_acumm = gr.Slider(
+                label="Grad accumulation steps:",
+                minimum=2,
+                maximum=128,
+                step=1,
+                value=args.grad_acumm,
             )
             progress_train = gr.Label(
                 label="Progress:"
@@ -201,10 +251,10 @@ with gr.Blocks() as demo:
             demo.load(read_logs, None, logs_tts_train, every=1)
             train_btn = gr.Button(value="Step 2 - Run the training")
 
-        def train_model(language, train_csv, eval_csv, num_epochs, batch_size, output_path, progress=gr.Progress(track_tqdm=True)):
+            def train_model(language, train_csv, eval_csv, num_epochs, batch_size, grad_acumm, output_path):
                 clear_gpu_cache()
 
-            config_path, original_xtts_checkpoint, vocab_file, exp_path, speaker_wav = train_gpt(language, num_epochs, batch_size, train_csv, eval_csv, output_path=output_path)
+                config_path, original_xtts_checkpoint, vocab_file, exp_path, speaker_wav = train_gpt(language, num_epochs, batch_size, grad_acumm, train_csv, eval_csv, output_path=output_path)
                 # copy original files to avoid parameters changes issues
                 os.system(f"cp {config_path} {exp_path}")
                 os.system(f"cp {vocab_file} {exp_path}")
@@ -295,6 +345,7 @@ with gr.Blocks() as demo:
                     eval_csv,
                     num_epochs,
                     batch_size,
+                    grad_acumm,
                     out_path,
                 ],
                 outputs=[progress_train, xtts_config, xtts_vocab, xtts_checkpoint, speaker_reference_audio],
@@ -320,12 +371,9 @@ with gr.Blocks() as demo:
                 outputs=[tts_output_audio, reference_audio],
             )
 
-
-
-if __name__ == "__main__":
     demo.launch(
         share=True,
-        debug=True,
-        server_port=PORT,
+        debug=False,
+        server_port=args.port,
         server_name="0.0.0.0"
     )