>>> data.nanquantile(0, axis=0)
tensor([0., 1., 6., 3.])

quantile(0) is the min value of your tensor and quantile(1) is the max value (quantile(0.5) is the median). The nan flavor ignores the nan values.

pytorch documentation