EXAQ: Exponent Aware Quantization For LLMs Acceleration

Documenti analoghi

• FP4 All the Way: Fully Quantized Training of LLMs
  di: Chmiel, Brian, et al.
  Pubblicazione: (2025)
• Scaling FP8 training to trillion-token LLMs
  di: Fishman, Maxim, et al.
  Pubblicazione: (2024)
• Normalized Architectures are Natively 4-Bit
  di: Fishman, Maxim, et al.
  Pubblicazione: (2026)
• Accurate Neural Training with 4-bit Matrix Multiplications at Standard Formats
  di: Chmiel, Brian, et al.
  Pubblicazione: (2021)
• Minimum Variance Unbiased N:M Sparsity for the Neural Gradients
  di: Chmiel, Brian, et al.
  Pubblicazione: (2022)