What was the loss values for the pertained models?

[xDuck]

I am training my own model and I was just curious what the expected final loss value is. I seem to be stagnating around 2.08 loss using all the default parameters.

[zhaoyun630]

I have the same problem. I'm training on 16k-LP7 dataset, the loss is converging at 2.7, which is very high to me.

[jmvalin]

The exact loss value does not matter actually. It's mostly dataset-dependent. It's usually somewhere between 2.0 and 3.0, but it could be outside that range on some data.