inferencerlabs

/

Qwen3.5-35B-A3B-MLX-9bit

like 5

It's the more precise bits per weight measurement.

It's the more precise bits per weight measurement.

sure but is there any specific architectural oomph that 9bit has on Apple Silicon over 8bit ? I mean 8 and 9 bit should be like twins running the same model really no ?

Yes, 9 bit would have higher precision available to each weight than 8.25bit.

Yes, 9 bit would have higher precision available to each weight than 8.25bit.

thanks , i was thinking it was something else , like to do with apple gpu architectural mapping benefit of some sort, but sure., better quality i get that too.

thanks for your uploads ! 💘