unsloth

/

MiniMax-M2-GGUF

like 71

Follow

Unsloth AI 11.1k

Is MXFP4 really better than Q4_K_XL or Q4_K_M? Or it's just because OpenAI used it?

It's worse than the k quants. But it has native hardware support and is fast and energy efficient.

It's worse than the k quants. But it has native hardware support and is fast and energy efficient.

So if my gpu is older than blackwell it's not worth using? Only blackwell has native fp4 support right?

Good idea we are going to do that pretty soon hopefully for our future quants. The quality will degrade slightly

It's worse than the k quants. But it has native hardware support and is fast and energy efficient.

So if my gpu is older than blackwell it's not worth using? Only blackwell has native fp4 support right?

Also works on old GPUS

Good idea we are going to do that pretty soon hopefully for our future quants. The quality will degrade slightly

It's worse than the k quants. But it has native hardware support and is fast and energy efficient.

So if my gpu is older than blackwell it's not worth using? Only blackwell has native fp4 support right?

Also works on old GPUS

But is the speed much slower than blackwell gpus?