vikigenius/textvinf

Variational Inference for text generation with Normalizing flows.

textvinf

Variational Inference for text with Normalizing Flows

Features

• Fully typed with annotations and checked with mypy, PEP561 compatible

Installation

• Create a conda environment
• Clone the repository
• Install poetry following the instructions from here
• Install using the command:

poetry install

Usage

To train download the ptblm dataset to the data folder or change the data location in config/ptblm_wae_nf.jsonnet

allennlp train -s models/ptblmwaenf --include-package textvinf config/ptblm_wae_nf.jsonnet

To generate sentences by sampling from the prior use:

textvinf generate models/ptblmwaenf/model.tar.gz --weights_file models/ptblmwaenf/model_state_epoch_47.th

License

MIT

Credits

This project was generated with wemake-python-package. Current template version is: f15a51da88c8a650d16e8f30fd30f9ddeda1aea8. See what is updated since then.