Speeding up Fastai Tabular with NumPy to avoid Submission Timeout Error

Since I am facing a timeout error as I submit for Jane Street competition, I thouth to use the notebook in https://muellerzr.github.io/fastblog/2020/04/22/TabularNumpy.html
to speed up my fastai changing from pandas to numpy.

Usually the lines here are enough to make inference:

to_tst = to.new(test_data)
tst_dl = dls.valid.new(to_tst)

preds, _ = learn.get_preds(dl=tst_dl)

which does not work when I use the numy code in the linked method.

What should I add to the code to run the inference?