how to interprete predictions of the model

[NxNiki]

I suppose the function memory is for running a model on test data. (How is it different from inference?)

brain_decoding/src/brain_decoding/trainer.py

Lines 396 to 397 in f6f15bd

	pred = output.cpu().detach().numpy()
	predictions = np.concatenate([predictions, pred], axis=0)

Should predictions be the probability each of the concepts is recalled? Why the values in each row does not sum to 1?
I also noticed that the number of rows of predictions is always 10000 regardless of the length of test data. Is the activation discretized into a fixed number of bins?