TFT predict() with output_dir() running out of memory on large validation sets

[pcgm-team]

It appears there is some sort of memory leak.

new_raw_predictions = model.predict(validation, mode="prediction", return_index = True, batch_size=128, output_dir = PATH)

[shellrazer]

same problem

[IlIlllIIllIIlll]

Me, too.

[zju-ys]

same problem