Closed anjani-dhrangadhariya closed 4 years ago
Yes, it is exactly what I put in the question. All the tensors should begin with a non-zero/False mask value.
@anjani-dhrangadhariya I also get the same error:
0%| | 3/62500 [00:01<5:18:57, 3.27it/s]
ValueError: mask of the first timestep must all be on
I am training BERT-CRF model. How to solve this error?
What exactly does this ValueError mean? This is the mask tensor (dim: [6, 512]) I am using
I assume it means that all the 6 tensors (dim 1) should begin with value 1 and not 0.