athomasson2/fine_tune_xtts repository overview

⁠Fine-Tune XTTS Docker Image

⁠Info about v2 and v4

This repository contains the athomasson2/fine_tune_xtts:v5 Docker image for fine-tuning XTTS models. This image is configured to work with CUDA for GPU acceleration.

⁠Prerequisites

• Docker installed on your machine
• NVIDIA Container Toolkit installed (for GPU support)

⁠Pulling the Docker Image

First, pull the Docker image from Docker Hub:

docker pull athomasson2/fine_tune_xtts:v5

⁠Running the Docker Image

⁠macOS

1. Open your terminal.
2. Navigate to the directory where you want to bind the training data (e.g., /Users/yourname/training).

cd /Users/yourname

1. Run the Docker container with GPU support and bind the training directory:

docker run --gpus all -it -v ${PWD}/training:/tmp/xtts_ft/ athomasson2/fine_tune_xtts:v5

⁠Windows

1. Open PowerShell.
2. Navigate to the directory where you want to bind the training data (e.g., C:\Users\yourname\Documents\FineTune-Xtts).

cd C:\Users\yourname\Documents\FineTune-Xtts

1. Run the Docker container with GPU support and bind the training directory:

docker run --gpus all -it -v ${PWD}\training:/tmp/xtts_ft/ athomasson2/fine_tune_xtts:v5

⁠Linux

1. Open your terminal.
2. Navigate to the directory where you want to bind the training data (e.g., /home/yourname/training).

cd /home/yourname

1. Run the Docker container with GPU support and bind the training directory:

docker run --gpus all -it -v ${PWD}/training:/tmp/xtts_ft/ athomasson2/fine_tune_xtts:v5

⁠for cpu only docker ( Will crash unless you go into your docker settings and bump your docker resource allocation limit all the way up. )

docker run -it -v ${PWD}/training:/tmp/xtts_ft/run/training athomasson2/fine_tune_xtts:v4_cpu

⁠Using the Docker Container

Once the container is running, it will automatically start the XTTS fine-tuning process. The training data should be placed in the bound training directory, which will be accessible inside the container at /tmp/xtts_ft/run/training.

⁠Accessing Container Logs

To access the logs and output of the container, simply use the terminal or PowerShell window where the container is running. All outputs and logs will be displayed there.

⁠Stopping the Container

To stop the Docker container, you can use the CTRL+C command in the terminal or PowerShell window where the container is running.

⁠Additional Information

For more details on how to use Docker, refer to the Docker documentation⁠.

For more details on NVIDIA Container Toolkit for GPU support, refer to the NVIDIA Container Toolkit documentation⁠.

⁠License

This project is licensed under the MIT License. See the LICENSE⁠ file for more details.

⁠Contact

⁠Closing the Docker Container for v4_cpu and v4 only

When you're done with the fine-tuning process, you can safely close the Docker container:

1. Press Ctrl+C in the terminal where the container is running.
2. This triggers a cleanup script that:
   • Condenses the trained model
   • Zips the dataset
   • Saves essential files (model.pth, config.json, vocab.json_, dataset.zip)
3. All processed files are saved in: ./training/Finished_model_files/

This ensures all your work is preserved and easily accessible after the container stops.

⁠INFO FOR old with v1 tag

This is a DOCKER image for fine tuning xtts on a nvida graphics card

docker run --gpus all -it -v ${PWD}\training:/tmp/xtts_ft/run/training athomasson2/fine_tune_xtts:v1

Is how the docker image is used on windows

That'll make it so that the training docker image folder points to your local folder on your computer named "training", then all the models generated in the program will be saved in that training folder on your computer.

if anything goes wrong the extra needed files you need to download outside of this image are on this huggingface repo

https://huggingface.co/drewThomasson/xtts_fine_tune_base_model_files⁠