Fix doc example: mask_time_indices (numpy) has no attribute 'to'

[ydshieh]

What does this PR do?

In speech to text models, there are doc examples

    >>> mask_time_indices = _compute_mask_indices((batch_size, sequence_length), mask_prob=0.2, mask_length=2)
    >>> with torch.no_grad():
    ...     outputs = model(input_values, mask_time_indices=mask_time_indices)

which gives numpy.ndarray, and the following line fails

transformers/src/transformers/models/wav2vec2/modeling_wav2vec2.py

Line 1481 in 19d37c2

mask_time_indices = mask_time_indices.to(torch.bool)

with AttributeError: 'numpy.ndarray' object has no attribute 'to'.

This PR add

mask_time_indices = torch.tensor(mask_time_indices, device=input_values.device, dtype=torch.bool)

in the examples to make them work.

Who can review?

@patrickvonplaten

[patrickvonplaten]

Thank you!