Implement Sliding Window Attention in PyTorch

Mediummodern-architecturesauto-graded

Restrict attention to a fixed local window around each token, reducing memory from O(n²) to O(n·w) for long sequences.

Solve it

Check your answer

The grader verifies properties of your implementation, so a correct solution written differently from ours still passes.

pip install torchleet

from torchleet import check
check("sliding-window-attention", create_sliding_window_mask, full_attention, sliding_window_attention)

Company tags

How these tags are sourced