Update xtts_demo.py added export model button

The exported model folder will contain these files

`dataset.zip`
`model.pth`
`config.json`
`vocab.json`

That will also remove the Optimizer State from the model making the exported model smaller and take up less space.

After all your just going to be using it for the inference anyway

TTS/demos/xtts_ft_demo/xtts_demo.py

@@ -14,6 +14,38 @@
 from TTS.tts.configs.xtts_config import XttsConfig
 from TTS.tts.models.xtts import Xtts
 
+def export_model(output_path):
+    try:
+        output_folder = os.path.join(output_path, "Finished_model_files")
+        os.makedirs(output_folder, exist_ok=True)
+
+        with zipfile.ZipFile(os.path.join(output_folder, "dataset.zip"), "w", zipfile.ZIP_DEFLATED) as zipf:
+            dataset_path = os.path.join(output_path, "dataset")
+            for root, dirs, files in os.walk(dataset_path):
+                for file in files:
+                    zipf.write(os.path.join(root, file),
+                               os.path.relpath(os.path.join(root, file), os.path.join(dataset_path, "..")))
+
+        search_path = os.path.join(output_path, "run", "training", "**", "best_model.pth")
+        model_path = max(glob.glob(search_path, recursive=True), key=os.path.getctime)
+
+        checkpoint = torch.load(model_path, map_location=torch.device("cpu"))
+        del checkpoint["optimizer"]
+        for key in list(checkpoint["model"].keys()):
+            if "dvae" in key:
+                del checkpoint["model"][key]
+
+        torch.save(checkpoint, os.path.join(output_folder, "model.pth"))
+
+        model_dir = os.path.dirname(model_path)
+        shutil.copy2(os.path.join(model_dir, "config.json"), output_folder)
+        shutil.copy2(os.path.join(model_dir, "vocab.json"), output_folder)
+
+        return f"Model exported successfully to {output_folder}"
+    except Exception as e:
+        return f"Export failed: {str(e)}"
+
+
 
 def clear_gpu_cache():
     # clear the GPU cache
@@ -384,6 +416,10 @@ def train_model(
                     progress_gen = gr.Label(label="Progress:")
                     tts_output_audio = gr.Audio(label="Generated Audio.")
                     reference_audio = gr.Audio(label="Reference audio used.")
+                # Button to export the fine-tuned model with optimizers removed
+                with gr.Column() as col4:
+                    export_btn = gr.Button(value="Export Fine-tuned Model")
+                    export_progress = gr.Label(label="Export Progress:")
 
             prompt_compute_btn.click(
                 fn=preprocess_dataset,
@@ -429,5 +465,11 @@ def train_model(
                 ],
                 outputs=[progress_gen, tts_output_audio, reference_audio],
             )
+
+            export_btn.click(
+                fn=export_model,
+                inputs=[out_path],
+                outputs=[export_progress],
+            )
 
     demo.launch(share=True, debug=False, server_port=args.port, server_name="0.0.0.0")