How a Bilingual LM Becomes Bilingual: Tracing Internal Representations with Sparse Autoencoders

Documenti analoghi

• Weight-based Analysis of Detokenization in Language Models: Understanding the First Stage of Inference Without Inference
  di: Kamoda, Go, et al.
  Pubblicazione: (2025)
• Do LLMs Need to Think in One Language? Correlation between Latent Language and Task Performance
  di: Ozaki, Shintaro, et al.
  Pubblicazione: (2025)
• Exclusive Unlearning
  di: Sasaki, Mutsumi, et al.
  Pubblicazione: (2026)
• TopK Language Models
  di: Takahashi, Ryosuke, et al.
  Pubblicazione: (2025)
• The Curse of Popularity: Popular Entities have Catastrophic Side Effects when Deleting Knowledge from Language Models
  di: Takahashi, Ryosuke, et al.
  Pubblicazione: (2024)