Test-Time Alignment of LLMs via Sampling-Based Optimal Control in pre-logit space

Documenti analoghi

• Evaluating Time-Series Training Dataset through Lens of Spectrum in Deep State Space Models
  di: Kanai, Sekitoshi, et al.
  Pubblicazione: (2024)
• Relative Density Ratio Optimization for Stable and Statistically Consistent Model Alignment
  di: Takahashi, Hiroshi, et al.
  Pubblicazione: (2026)
• Post-pre-training for Modality Alignment in Vision-Language Foundation Models
  di: Yamaguchi, Shin'ya, et al.
  Pubblicazione: (2025)
• Long-term Safe Reinforcement Learning with Binary Feedback
  di: Wachi, Akifumi, et al.
  Pubblicazione: (2024)
• Transfer Learning with Pre-trained Conditional Generative Models
  di: Yamaguchi, Shin'ya, et al.
  Pubblicazione: (2022)