Fix PyTorch warning

[junaid-hasan]

Summary

This PR resolves several UserWarning messages from PyTorch by

• updating deprecated function calls in the custom Adam optimizer located in src/optim.py.
• updating function calls in src/model/transformer.py and src/model/lstm.py.
• updating .gitignore to include a dump directory for saving temporary training files.

This is a non-functional change that ensures forward compatibility with future PyTorch versions and cleans up the logs.

Changes

The following deprecated function signatures were updated:

• add_(Number, Tensor) is now add_(Tensor, *, alpha=Number)
• addcmul_(Number, Tensor, Tensor) is now addcmul_(Tensor, Tensor, *, value=Number)
• addcdiv_(Number, Tensor, Tensor) is now addcdiv_(Tensor, Tensor, *, value=Number)
• replace .byte() with .bool() in masking
• Added dump directory to .gitignore

Example:

Before:

• p.data.addcdiv_(-step_size, exp_avg, denom)
• generated[-1].masked_fill_(unfinished_sents.byte(), self.eos_index)

After:

• p.data.addcdiv_(exp_avg, denom, value=-step_size)
• generated[-1].masked_fill_(unfinished_sents.bool(), self.eos_index)

Context

This change is motivated by an API evolution in PyTorch, as discussed in the official PyTorch repository.

• Related GitHub Issue: pytorch/pytorch#32861
• PyTorch Discussions overload of add_ , addcmul_, addcdiv_ is deprecated
• and dtype torch.uint8 is now deprecated