strix-halo-kernels / test_flash.py

Commit History

Add Triton flash attention (2-3x over AOTriton, 31x less peak memory)
b3e26c5
verified

axjns commited on