Decoupling the "What" and "Where" With Polar Coordinate Positional Embeddings

Documenti analoghi

• Mixture of Sparse Attention: Content-Based Learnable Sparse Attention via Expert-Choice Routing
  di: Piękos, Piotr, et al.
  Pubblicazione: (2025)
• Recurrent Complex-Weighted Autoencoders for Unsupervised Object Discovery
  di: Gopalakrishnan, Anand, et al.
  Pubblicazione: (2024)
• MoEUT: Mixture-of-Experts Universal Transformers
  di: Csordás, Róbert, et al.
  Pubblicazione: (2024)
• SwitchHead: Accelerating Transformers with Mixture-of-Experts Attention
  di: Csordás, Róbert, et al.
  Pubblicazione: (2023)
• MrT5: Dynamic Token Merging for Efficient Byte-level Language Models
  di: Kallini, Julie, et al.
  Pubblicazione: (2024)