Merge branch 'erew123:alltalkbeta' into alltalkbeta

finetune.py

@@ -1598,15 +1598,17 @@ if __name__ == "__main__":
                     )
                     whisper_model = gr.Dropdown(
                         label="Whisper Model",
-                        value="large-v2",
+                        value="distil-large-v3",
                         choices=[
-                            "large-v3",
-                            "large-v2",
-                            "large",
-                            "medium",
-                            "small"
+                            ("large-v3", "large-v3"),
+                            ("large-v2", "large-v2"),
+                            ("large", "large"),
+                            ("medium", "large"),
+                            ("small", "large"),
+                            ("distil-large-v3 (en only)", "distil-large-v3"),
+                            ("distil-large-v2 (en only)", "distil-large-v2"),
                         ],
-                        scale=1,
+                        scale=2,
                     )
 
                     lang = gr.Dropdown(
@@ -1644,7 +1646,7 @@ if __name__ == "__main__":
                         minimum=5,  # Minimum value
                         maximum=95,  # Maximum value
                         step=1,  # Increment step
-                        scale=2,
+                        scale=1,
                     )
                     create_bpe_tokenizer = gr.Checkbox(
                         label='BPE Tokenizer',
@@ -1719,7 +1721,7 @@ if __name__ == "__main__":
                     3. Provide at least 2 minutes of audio, preferably 5-10 minutes for better results.<br>
                     4. Maximum Audio Length will ensure the audio samples do not exceed the set value in seconds. The AI model can only process so much audio per audio sample, per epoch.<br>
                     5. Enter a unique Project Name if you wish to create a dedicated training folder. You will need to specify this name as you move through the interfaces and click the Refresh buttons where needed.<br>
-                    6. Select the Whisper model (large-v2 recommended for English use cases).<br>
+                    6. Select the Whisper model (distil-large-v3 recommended for English use cases).<br>
                     7. Choose your Dataset Language.<br>
                     8. Adjust the Evaluation Split percentage (default 15% is suitable for most cases).<br>
                     9. Optionally enable the BPE Tokenizer for custom tokenization, which can improve the model's ability to handle unique words, accents, or languages by creating a vocabulary tailored to your specific dataset. This is especially useful for non-standard speech patterns or languages with complex morphology.<br>