Trajectory as the Teacher: Few-Step Discrete Flow Matching via Energy-Navigated Distillation

Documenti analoghi

• FS-DFM: Fast and Accurate Long Text Generation with Few-Step Diffusion Language Models
  di: Monsefi, Amin Karimi, et al.
  Pubblicazione: (2025)
• DACA-GRPO: Denoising-Aware Credit Assignment for Reinforcement Learning in Diffusion Language Models
  di: Monsefi, Amin Karimi, et al.
  Pubblicazione: (2026)
• M2R2: Mixture of Multi-Rate Residuals for Efficient Transformer Inference
  di: Bhendawade, Nikhil, et al.
  Pubblicazione: (2025)
• Mirror Speculative Decoding: Breaking the Serial Barrier in LLM Inference
  di: Bhendawade, Nikhil, et al.
  Pubblicazione: (2025)
• Speculative Streaming: Fast LLM Inference without Auxiliary Models
  di: Bhendawade, Nikhil, et al.
  Pubblicazione: (2024)