Block-Wise Differentiable Sinkhorn Attention: Tail-Refinement Gradients with a Gap-Aware Dustbin Bridge

Documenti analoghi

• LoSA: Locality Aware Sparse Attention for Block-Wise Diffusion Language Models
  di: Xi, Haocheng, et al.
  Pubblicazione: (2026)
• Improving Block-Wise LLM Quantization by 4-bit Block-Wise Optimal Float (BOF4): Analysis and Variations
  di: Blumenberg, Patrick, et al.
  Pubblicazione: (2025)
• Sinkhorn Distance Minimization for Knowledge Distillation
  di: Cui, Xiao, et al.
  Pubblicazione: (2024)
• EBFT: Effective and Block-Wise Fine-Tuning for Sparse LLMs
  di: Guo, Song, et al.
  Pubblicazione: (2024)
• Optimizing Mixture of Block Attention
  di: Xiao, Guangxuan, et al.
  Pubblicazione: (2025)