I do hope they provide optimized A3B quants--that's been the sweet spot for usable local inference for me, at least.