loading
loading
A language model that generates text by iteratively denoising a whole sequence at once, instead of one token left-to-right.
Borrowing from image diffusion, these models start from masked or noisy tokens and refine the entire output over several passes, rather than the autoregressive token-by-token approach of standard LLMs. The appeal is parallelism — you can produce many tokens per step and revise earlier ones, which can mean much faster generation and built-in self-correction. They're still early and less proven than transformers for general text, but worth watching as a speed play for code and structured output.
Plainly
Think of Diffusion LLM as the brain part that guesses or decides. A language model that generates text by iteratively denoising a whole sequence at once, instead of one token left-to-right.
In practice
Use it when model choice, prompts, latency, cost, or quality affect the product result. In practice, define the owner, input, output, and failure mode before you rely on it.