Right.I was thinking about it, you still need batch refill, however, Apple Core ML tools were failing for attention activations quantization. Long context, pre

Right.I was thinking about it, you still need batch refill, however, Apple Core ML tools were failing for attention activations quantization. Long context, pre-fill is still compute bound.