espnet2.asr.transducer.rnnt_multi_blank.utils.rnnt_helper.copy_data_1d

source

espnet2.asr.transducer.rnnt_multi_blank.utils.rnnt_helper.copy_data_1d(source: Tensor, dest: Tensor, idx: int)

Copies a single element from a source tensor to a destination tensor at the specified index.

This function is intended for use in CUDA kernels to facilitate data manipulation between tensors. It assumes that both source and dest are 1-dimensional tensors and that the provided index is within the bounds of these tensors.

• Parameters:
  • source (torch.Tensor) – The source tensor from which to copy data. It must be a 1-dimensional tensor.
  • dest (torch.Tensor) – The destination tensor where data will be copied. It must also be a 1-dimensional tensor.
  • idx (int) – The index at which to copy the data from the source tensor to the destination tensor.

Examples

>>> import torch
>>> source_tensor = torch.tensor([1.0, 2.0, 3.0])
>>> dest_tensor = torch.zeros(3)
>>> copy_data_1d(source_tensor, dest_tensor, 1)
>>> print(dest_tensor)  # Output: tensor([0.0, 2.0, 0.0])

NOTE

This function should be called within a CUDA kernel and is not intended for direct invocation in Python code.