XiangLi1999/Diffusion-LM
Diffusion-LM is a diffusion-model-based approach to controllable text generation that learns to denoise sequences of embeddings to produce text conditioned on classifier-guided constraints.
Not currently ranked — collecting fresh signals.
star history
This repository implements a diffusion model architecture for text generation, replacing autoregressive decoding with an iterative denoising process over continuous text embeddings. The model trains on paired datasets (E2E, ROCstory) using a transformer backbone and supports controllable generation by training a classifier (e.g., syntactic parser) to guide the diffusion sampling process toward desired attributes.
Frequently asked
- What is XiangLi1999/Diffusion-LM?
- Diffusion-LM is a diffusion-model-based approach to controllable text generation that learns to denoise sequences of embeddings to produce text conditioned on classifier-guided constraints.
- Is Diffusion-LM open source?
- Yes — XiangLi1999/Diffusion-LM is open source, released under the Apache-2.0 license.
- What language is Diffusion-LM written in?
- XiangLi1999/Diffusion-LM is primarily written in Python.
- How popular is Diffusion-LM?
- XiangLi1999/Diffusion-LM has 1.2k stars on GitHub.
- Where can I find Diffusion-LM?
- XiangLi1999/Diffusion-LM is on GitHub at https://github.com/XiangLi1999/Diffusion-LM.