AGoQ: Activation and Gradient Quantization for Memory-Efficient Distributed Training of LLMs

Documenti analoghi

• Parm: Efficient Training of Large Sparsely-Activated Models with Dedicated Schedules
  di: Pan, Xinglin, et al.
  Pubblicazione: (2024)
• ZipCCL: Efficient Lossless Data Compression of Communication Collectives for Accelerating LLM Training
  di: Lin, Wenxiang, et al.
  Pubblicazione: (2026)
• HierMoE: Accelerating MoE Training with Hierarchical Token Deduplication and Expert Swap
  di: Lin, Wenxiang, et al.
  Pubblicazione: (2025)
• Efficient MoE Inference with Fine-Grained Scheduling of Disaggregated Expert Parallelism
  di: Pan, Xinglin, et al.
  Pubblicazione: (2025)
• QSync: Quantization-Minimized Synchronous Distributed Training Across Hybrid Devices
  di: Zhao, Juntao, et al.
  Pubblicazione: (2024)