Nexesenex

/

MIstral-QUantized-70b_Miqu-1-70b-iMat.GGUF

like 67

That's because such requants would be useless for inference.

Miqudev released only a q5_K_M as the best quant available, which I used to make all these requants.

Any requant superieur to Q4_K_S (Miqudev also provided a Q4_K_M quant) would not be better than what Miqudev provided, because these requants would be baed on his Q5_K_M.

ah okay, if initial model was just Q5_K_M, then by making quant higher you will not make the model perform better. That's obvious, my bad, thanks! 🙏