# Training

Important!

Training is only for NVDIA GPUs, If you don’t have a compatible GPU, the Training tab will be disabled.

If you don’t have an NVIDIA GPU, you can try Applio Colab or Applio No UI Colab. Check the other alternatives section.

Step 1: Dataset Preparation

Upload your audio in .wav format using the Dataset Maker (if it is a single audio) or setup it manually (various audios) to applio/assets/datasets creating inside a folder for the program to read it.

don't know how to create a dataset?, check the dataset section

Step 2: Dataset Processing

Once the model is named and the dataset selected press "Prepocess Dataset" and wait for the message in the CMD.

Make sure your dataset consists of single audio file or if you split it, ensure that each audio segment has a duration of 10 to 15 seconds per audio.

you can select one of the 3 available frequencies according to the audio (32k, 40k, 48k), this will help to avoid filtering out more artifacts or background noise.
don't know how to check sample rate?, check the sample rate section

Step 3: Feature Extraction

Select an F0 method that suits your needs.
(optional) modify Hop lenght, lower value, higher smoothness of pitch change but slower training and vice versa.
(optional) select the Embedder model (hubert or contentvec)

Step 4: Model Training

Configure the training parameters according to your needs.

Save Every Epoch: Set this value between 10 and 50 to determine how often the model's state is saved during training.
Total Epochs: The number of epochs needed varies based on your dataset. Monitor progress using TensorBoard; typically, models perform well around 200-400 epochs.
Batch Size: Adjust based on your GPU's VRAM. For 8 GB VRAM, use a batch size between 6 and 8. Consider CUDA cores when experimenting with higher batch sizes.

# Other Options

Pitch Guidance: Gives variation of pitch.
Pretrained: Uses the RVC pretrained.
Save Only Latest: Save a single D/G file with information.
Save Every Weights: Save the weights of the model when a cycle of 'Save Every Epoch' is completed.
Custom Pretrained: Uses the Custom Pretrained that are loaded.
GPU Settings: Allows to choose GPUs (only for users who have more than one GPU).
Overtraining Detector: Mark it only if you will train for more than 200 epochs.
Overtraining Threshold: Set the maximum number of epochs you want your model to stop training if no improvement is detected.

Mark the Save Only Latest option before training to prevent it from filling up your storage.

Once configured, press 'Start training' to start the process, everything is registered in the CMD.

don't know how to use Tensorboard for correct training? Check out the Tensorboard section

Final Step: Model Saving and Index File Generation

Once training is completed, generate the index file by clicking the "Train Feature Index" button.
Your trained model is located in the logs/model folder, and the .pth files are in the logs/zips folder.

Now you can export your trained model directly from the Applio interface, go to the Export Model section in the train tab, click on the Refresh button and select the pth and the added index of the model to export.

Optional Step: Resume training

Open Applio if you have closed it.
Then, in the Applio interface, input your model name, use the same sample rate, and proceed to the last part of the train tab. Set the same batch size, pretrained (if you used) and increase the number of epochs you want to train.
Once configured, press 'Start training' to start the process, everything is registered in the CMD.