The sycl_ext_oneapi_device_architecture is currently experimental extension while actual products such as Pytorch rely on it in the shipped products. Can this extension be promoted from experimental to supported?
See: https://github.com/intel/torch-xpu-ops/blob/release/2.14/src/ATen/native/transformers/xpu/flash_attn/sycltla/mha_bwd.cpp#L75
CC: @gmlueck, @EikanWang, @guangyey
The sycl_ext_oneapi_device_architecture is currently experimental extension while actual products such as Pytorch rely on it in the shipped products. Can this extension be promoted from experimental to supported?
See: https://github.com/intel/torch-xpu-ops/blob/release/2.14/src/ATen/native/transformers/xpu/flash_attn/sycltla/mha_bwd.cpp#L75
CC: @gmlueck, @EikanWang, @guangyey