Shark-NLP/DiffuSeq
DiffuSeq is a conditional diffusion language model for sequence-to-sequence text generation, trained end-to-end in a classifier-free manner.
Not currently ranked — collecting fresh signals.
star history
DiffuSeq implements a diffusion-based approach to sequence-to-sequence text generation, where a model learns to gradually denoise text sequences conditioned on input. The project provides an accelerated version (DiffuSeq-v2) that achieves 800x faster sampling while maintaining generation quality. It supports various text generation tasks including summarization, paraphrase, and dialogue generation through a unified conditional diffusion framework.
Frequently asked
- What is Shark-NLP/DiffuSeq?
- DiffuSeq is a conditional diffusion language model for sequence-to-sequence text generation, trained end-to-end in a classifier-free manner.
- Is DiffuSeq open source?
- Yes — Shark-NLP/DiffuSeq is open source, released under the MIT license.
- What language is DiffuSeq written in?
- Shark-NLP/DiffuSeq is primarily written in Python.
- How popular is DiffuSeq?
- Shark-NLP/DiffuSeq has 838 stars on GitHub.
- Where can I find DiffuSeq?
- Shark-NLP/DiffuSeq is on GitHub at https://github.com/Shark-NLP/DiffuSeq.