Container build dependency on accelerators (requires TPU VM)

[ultrons]

https://github.com/google/maxdiffusion/blob/inf_mlperf/docker/sdxl_inference/model_loader.py
is executed the dockerfile and requires the build to be executed on TPU VM.
Can we replace this with xAOT to eliminate this dependency?

[entrpn]

@ultrons it is not a requirement to build Docker on a TPU VM. When this gets loaded, it can use cpu, see this line.

[entrpn]

@ultrons does this answer your question?