Marianne Arriola
@mariannearr
Hope you can catch our #NeurIPS2025 poster (#3500) tonight at 4:30PM! We discuss limitations of using decoder-only transformers for diffusion LMs and why encoder-decoders are better for faster inference.
x.com/mariannearr/st…
x.com/mariannearr/st…
Marianne Arriola@mariannearr · Oct 28, 2025🚨In our NeurIPS paper, we bring encoder-decoders back.. for diffusion language models!
⚡️Encoder-decoders make diffusion sampling fast: a small (fast) decoder denoises tokens progressively and a large (slower) encoder represents clean context.
⚡️Encoder-decoders make diffusion sampling fast: a small (fast) decoder denoises tokens progressively and a large (slower) encoder represents clean context.
0:31 0 46