Garbage Attention in Large Language Models: BOS Sink Heads and Sink-aware Pruning

Documenti analoghi

• EpiCaR: Knowing What You Don't Know Matters for Better Reasoning in LLMs
  di: Yeom, Jewon, et al.
  Pubblicazione: (2026)
• Hallucination as Commitment Failure: Larger LLMs Misfire Despite Knowing the Answer
  di: Yeom, Jewon, et al.
  Pubblicazione: (2026)
• Efficient Epistemic Uncertainty Estimation for Large Language Models via Knowledge Distillation
  di: Park, Seonghyeon, et al.
  Pubblicazione: (2026)
• From Noise to Diversity: Random Embedding Injection in LLM Reasoning
  di: Kim, Heejun, et al.
  Pubblicazione: (2026)
• Prefixing Attention Sinks can Mitigate Activation Outliers for Large Language Model Quantization
  di: Son, Seungwoo, et al.
  Pubblicazione: (2024)