PAT: a new algorithm for all-gather and reduce-scatter operations at scale

Documenti analoghi

• Demystifying NCCL: An In-depth Analysis of GPU Communication Protocols and Algorithms
  di: Hu, Zhiyi, et al.
  Pubblicazione: (2025)
• Configurable Non-uniform All-to-all Algorithms
  di: Fan, Ke, et al.
  Pubblicazione: (2024)
• Optimal, Non-pipelined Reduce-scatter and Allreduce Algorithms
  di: Träff, Jesper Larsson
  Pubblicazione: (2024)
• PAT: Accelerating LLM Decoding via Prefix-Aware Attention with Resource Efficient Multi-Tile Kernel
  di: Yi, Jinjun, et al.
  Pubblicazione: (2025)
• Scaling All-to-all Operations Across Emerging Many-Core Supercomputers
  di: Kinkead, Shannon, et al.
  Pubblicazione: (2026)