Spectral Lens: Activation and Gradient Spectra as Diagnostics of LLM Optimization

Documenti analoghi

• Memory Inception: Latent-Space KV Cache Manipulation for Steering LLMs
  di: Liu, Andy Zeyi, et al.
  Pubblicazione: (2026)
• To Clip or not to Clip: the Dynamics of SGD with Gradient Clipping in High-Dimensions
  di: Marshall, Noah, et al.
  Pubblicazione: (2024)
• Eigenvalue distribution of the Neural Tangent Kernel in the quadratic scaling
  di: Benigni, Lucas, et al.
  Pubblicazione: (2025)
• High-Dimensional Privacy-Utility Dynamics of Noisy Stochastic Gradient Descent on Least Squares
  di: Lin, Shurong, et al.
  Pubblicazione: (2025)
• Dynamics of Stochastic Momentum with Sparse Updates in High Dimensions
  di: Everett, Katie, et al.
  Pubblicazione: (2026)