Arctic Long Sequence Training: Scalable And Efficient Training For Multi-Million Token Sequences

Documenti analoghi

• Shift Parallelism: Low-Latency, High-Throughput LLM Inference for Dynamic Workloads
  di: Hidayetoglu, Mert, et al.
  Pubblicazione: (2025)
• SwiftKV: Fast Prefill-Optimized Inference with Knowledge-Preserving Model Transformation
  di: Qiao, Aurick, et al.
  Pubblicazione: (2024)
• Arctic Inference with Shift Parallelism: Fast and Efficient Open Source Inference System for Enterprise AI
  di: Rajbhandari, Samyam, et al.
  Pubblicazione: (2025)
• MoE-Prefill: Zero Redundancy Overheads in MoE Prefill Serving
  di: Su, Zhaoyuan, et al.
  Pubblicazione: (2026)
• STUN: Structured-Then-Unstructured Pruning for Scalable MoE Pruning
  di: Lee, Jaeseong, et al.
  Pubblicazione: (2024)