Combining Domain and Alignment Vectors to Achieve Better Knowledge-Safety Trade-offs in LLMs

Documenti analoghi

• A Deep Dive into the Trade-Offs of Parameter-Efficient Preference Alignment Techniques
  di: Thakkar, Megh, et al.
  Pubblicazione: (2024)
• LLMs Can't Play Hangman: On the Necessity of a Private Working Memory for Language Agents
  di: Baldelli, Davide, et al.
  Pubblicazione: (2026)
• Probabilistic Calibration Is a Trainable Capability in Language Models
  di: Baldelli, Davide, et al.
  Pubblicazione: (2026)
• Fundamental Safety-Capability Trade-offs in Fine-tuning Large Language Models
  di: Chen, Pin-Yu, et al.
  Pubblicazione: (2025)
• Exploring Quantization for Efficient Pre-Training of Transformer Language Models
  di: Chitsaz, Kamran, et al.
  Pubblicazione: (2024)