>>105635469
I already tried it and it was worse than using a 30B model at a normal quant.