Another improved quantization method from QTIP team?

[Originalimoc]

I just noticed this, https://github.com/Cornell-RelaxML/yaqa-quantization. How does this diff from and fits into exl3 one?

[Vhallo]

Unlikely to be feasible in the near future as YAQA quantization requires a massive amount of compute = expensive.