Skip to content

Installation

This package requires Python 3.10 or later and PyTorch 2.5 or later for execution. PyTorch must be installed according to your CUDA version. In normal mode, the model is optimized for a GPU, and a GPU with at least 8GB of VRAM is recommended. While it can run on a CPU, expect long execution times. In efficient mode, it is designed to provide fast inference even on a CPU.

from PYPI

pip install yomitoku

using uv

This repository uses the package management tool uv. After installing uv, clone the repository and execute the following commands:

uv sync

Inferencing with ONNX Runtime on a GPU

uv sync --extra gpu

When using uv, you need to modify the following part of the pyproject.toml file to match your CUDA version. By default, PyTorch compatible with CUDA 12.4 will be downloaded.

[[tool.uv.index]]
name = "pytorch-cuda124"
url = "https://download.pytorch.org/whl/cu124"
explicit = true

Using docker

A Dockerfile is provided in the root of the repository, which you are welcome to use.

docker build -t yomitoku .

docker run -it --gpus all -v $(pwd):/workspace --name yomitoku yomitoku /bin/bash

docker run -it -v $(pwd):/workspace --name yomitoku yomitoku /bin/bash

Using YomiToku in Offline Environments

YomiToku automatically downloads the model from Hugging Face Hub on its first run. An internet connection is required at that time but, by manually pre-downloading the model using the following command, you can prepare YomiToku to run in environments without internet access:

download_model

By placing the downloaded repository folder KotaroKinoshita in the current directory at runtime, the local repository model will be loaded and executed without any internet connection.