3x3 convolution with padding

Args: index: (b t), index of the masked point len_out: length of mask reverse: reverse the output mask (Default: False)

Compute the multiclass accuracy w.r.t. groundtruth :param preds: Float array representing the predictions, dimension (N,) :param truths:

Returns the activation function corresponding to `activation`

Computes the accuracy over the k top predictions for the specified values of k

Args: in_size: input dimension hidden_size: hidden layer dimension dropout: dropout probability O

add position information to the input x x: B, T, C position: T, C mask: B, T

Args: start_index: (b t), the first true value end_index: (b t), the last true value len_out: length of mask rev

1x1 convolution

FP16-compatible function that fills a tensor with -inf.