RND1 Diffusion Language Model

Generate text using a diffusion-based language model that refines all tokens simultaneously through iterative denoising steps.
Generation Mode

Task: Q&A format | Completion: Text continuation

Display output at each diffusion step (slower but shows the process)
16 256
32 512
0.1 2
0 100
0 1
0 100000

How it works

Diffusion models generate text differently than standard language models:

  1. Initialize all tokens as noise/masks simultaneously
  2. Iteratively denoise and refine all tokens together
  3. After N steps, the output converges to coherent text

With live generation enabled, you can watch the text improve step by step.
Examples
Prompt Generation Mode Diffusion Steps Max New Tokens Temperature Top-K Top-P (Nucleus) Random Seed Show Live Generation