MXFP4 quants for new models?

#2024
by n0head - opened

Hello. I wonder if there is a reason you're not making mxfp4 quants for new model and if there is a possibility you'll start making them in the future for RTX 5xxx series card users?

We are only providing mxfp4 quants for GPT OSS based modes and other models that got uploaded in mxfp4. It’s a manual process to add mxfp4 quants when queueing a model so sometimes we forget in which case you can remind us. The mxfp4 GGUF mixture is controversial and many claims it is terrible even when used for mxfp4 based source models and based on my own experience they are right.

Thanks for the reply.

n0head changed discussion status to closed

Sign up or log in to comment