[Feat]: Minimaxh3 support?

[alperktt]

Describe your use-case.

Minimaxh3 support? int8 convrot diffusion and text encoder

What would you like to see as a solution?

Minimaxh3 support? int8 convrot diffusion and text encoder

Have you considered alternatives? List them here.

aitoolkit didnt worked

[wogam]

Bumping this as OT is basically the only good implementation of LoRA training.

[dxqb]

there is some infrastructure being built before supporting huge video models, because nobody is being helped by supporting a model but then require 128 GB RAM at today's prices.

LTX is the prototype. if you want to see this happen join our discord and help testing when I post those PRs