hello,i found my loss is always nan.

[Kissacat]

hello, i run your code to train the model. i found my loss including test-loss is always Nan. The output of the model is 0 (dim = 2). "output = F.log_softmax(y_hat+1e-10)" (in line 154 in deepspeech_main.py). It doesn't work that y_hat plus with [eps.[]