That's a thing
https://hackernoon.com/what-is-a-diffusion-llm-and-why-does-it-matter
Nice because generation can be more parallel
Discussion
Also has implications for hallucination minimization and structured output enforcement, interesting
That's a thing
https://hackernoon.com/what-is-a-diffusion-llm-and-why-does-it-matter
Nice because generation can be more parallel
Also has implications for hallucination minimization and structured output enforcement, interesting