Latent Diffusion for Language Generation
This addresses the challenge of adapting diffusion models to language generation for researchers and practitioners, offering a novel hybrid approach that is incremental but improves upon existing methods.
The paper tackles the problem of applying diffusion models to discrete language data by proposing a method that uses encoder-decoder language models to learn high-quality language autoencoders, then trains continuous diffusion models in the latent space, enabling effective unconditional, class-conditional, and sequence-to-sequence language generation. The result shows that these latent language diffusion models are significantly more effective than previous diffusion language models across multiple diverse datasets.
Diffusion models have achieved great success in modeling continuous data modalities such as images, audio, and video, but have seen limited use in discrete domains such as language. Recent attempts to adapt diffusion to language have presented diffusion as an alternative to existing pretrained language models. We view diffusion and existing language models as complementary. We demonstrate that encoder-decoder language models can be utilized to efficiently learn high-quality language autoencoders. We then demonstrate that continuous diffusion models can be learned in the latent space of the language autoencoder, enabling us to sample continuous latent representations that can be decoded into natural language with the pretrained decoder. We validate the effectiveness of our approach for unconditional, class-conditional, and sequence-to-sequence language generation. We demonstrate across multiple diverse data sets that our latent language diffusion models are significantly more effective than previous diffusion language models.