LEAP: Layer-wise Exit-Aware Pretraining for Efficient Transformer Inference

Documenti analoghi

• SURGE: SuperBatch Unified Resource-efficient GPU Encoding for Heterogeneous Partitioned Data
  di: Kapadia, Shashank, et al.
  Pubblicazione: (2026)
• HELIOS: Adaptive Model And Early-Exit Selection for Efficient LLM Inference Serving
  di: Kumar, Avinash, et al.
  Pubblicazione: (2025)
• DEL: Context-Aware Dynamic Exit Layer for Efficient Self-Speculative Decoding
  di: Zarch, Hossein Entezari, et al.
  Pubblicazione: (2025)
• On the Effect of Uncertainty on Layer-wise Inference Dynamics
  di: Kim, Sunwoo, et al.
  Pubblicazione: (2025)
• LayerSkip: Enabling Early Exit Inference and Self-Speculative Decoding
  di: Elhoushi, Mostafa, et al.
  Pubblicazione: (2024)