From Attention Sensitivity to Layer Role: Revisiting Mixed-Precision Quantization of Transformers

arXiv:2609.34866 · cs.LG, cs.AI · Submitted 2026-09-28 · Read on arXiv

cs.LG, cs.AI

Submitted: 2026-09-28

Updated: 2026-09-28

Terminology

• Quantization — 1× in this paper · explained in 12 episodes
• Transformer — 1× in this paper · explained in 2 episodes

Sources

• Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation
• Mistral 7B
• Q-Strata: Hierarchical Bit Allocation for Mixed-Precision Quantization of Mixture-of-Experts LLMs
• AWP: Activation-Aware Weight Pruning and Quantization with Projected Gradient Descent
• OmniQuant: Omnidirectionally Calibrated Quantization for Large Language Models
• QuIP#: Even Better LLM Quantization with Hadamard Incoherence and Lattice Codebooks

Related papers

• Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
• AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
• Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
• Convergence issues in Relational Concept Analysis based on AOC-posets
• Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
• Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks