can the output model be transform to onnx format?

[dlkht]

as title

[hzhwcmhf]

Not supported yet.
I have a plan to achieve quick and effective deployment but it is not my highest priority (I'm still pursuing higher quality). PR is welcome.

[dlkht]

thanks for your answer.
and can it be run in android platform? (inference model)

[hzhwcmhf]

@dlkht I am afraid it is not in my future plan.

I want to mention that NAT is mainly designed to utilize the parallel computing ability of GPU, so it may not bring such a big speedup on CPU/mobile device. (See Table1 in https://arxiv.org/pdf/2205.10577.pdf)