4 ms·

I worked on it for a more specialized task (query rewriting). It’s blazing fast. A lot of inference code is set up for autoregressive decoding now. Diffusion i

I worked on it for a more specialized task (query rewriting). It’s blazing fast. A lot of inference code is set up for autoregressive decoding now. Diffusion is less mature. Not sure if Ollama or llama cpp support it.

stavros 8mo ago
How was the quality?
- janalsncm 8mo ago
  Quality was about the same. I will say it was a pain to train since it isn’t as popular and there isn’t out of the box support.
  - stavros 8mo ago
    Interesting, thanks! That's pretty cool though!
philipportner 8mo ago
Did you publish anything you could link wrt. query rewriting?