Use lru_cache to cache unique (ph, pw, device) position grids, avoiding
redundant computation when multiple images share the same patch
dimensions. Cache persists across forward passes.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
- Replace manual attention with F.scaled_dot_product_attention
- Use repeat_interleave instead of meshgrid for position computation
- Build image_ids efficiently with repeat_interleave instead of F.pad
- Remove unused Rearrange import
~56% speedup (91ms -> 58ms on 512 variable-sized images)
Numerically equivalent (max diff ~5e-4, within flash attention tolerance)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-authored-by: Claude <noreply@anthropic.com>
