Divergent Token Metrics: Measuring degradation to prune away LLM components -- and optimize quantization

Documenti analoghi

• T-FREE: Subword Tokenizer-Free Generative LLMs via Sparse Representations for Memory-Efficient Embeddings
  di: Deiseroth, Björn, et al.
  Pubblicazione: (2024)
• LIME: Making LLM Data More Efficient with Linguistic Metadata Embeddings
  di: Sztwiertnia, Sebastian, et al.
  Pubblicazione: (2025)
• Measuring and Guiding Monosemanticity
  di: Härle, Ruben, et al.
  Pubblicazione: (2025)
• AtMan: Understanding Transformer Predictions Through Memory Efficient Attention Manipulation
  di: Deiseroth, Björn, et al.
  Pubblicazione: (2023)
• SCAR: Sparse Conditioned Autoencoders for Concept Detection and Steering in LLMs
  di: Härle, Ruben, et al.
  Pubblicazione: (2024)