Meta's Llama 3.2 instruct models for ExecuTorch XNNPACK. 8da4w int4, round-to-nearest. One .pte per context window.