Sampling from a SVGP

[Anjum48]

I'm using the standard SVGP example from here.

What is the correct way to sample from a fitted model?

In the example under making predictions, the x_batch (from the test set) has a shape of [248, 18] (batch size = 248, features = 18).

I want to sample from the model given some x data. This is what I'm doing now:

samples = preds.sample(sample_shape=torch.Size([1000]))

Which gives an output shape of [1000, 248]. I'm looking for an expected shape of [1000, 18] - any ideas where I'm going wrong?