9 papers
Anti-Shortcut Distillation via Temporal Negative Knowledge Transfer
Syed Muhammad Raza, Omer Tariq, Jeongbae Son
Knowledge distillation (KD) trains a compact student by attracting it towards a converged teacher. It is silent about which directions the teacher itself learned to suppress: repul…
Zeta: Dual Whitening for Matrix Optimization via Coordinate-Adaptive Preconditioning
Kaiwen Chen, Shuhai Zhang, Zimo Liu +7
Large-scale neural network training increasingly relies on matrix-aware optimizers that exploit the structure of weight parameters beyond element-wise adaptation. However, existing…
Attention Sink in Transformers: A Survey on Utilization, Interpretation, and Mitigation
Zunhai Su, Hengyuan Zhang, Wei Wu +24
As the foundational architecture of modern machine learning, Transformers have driven remarkable progress across diverse AI domains. Despite their transformative impact, a persiste…
Residual Stream Duality in Modern Transformer Architectures
Yifan Zhang
Recent work has made clear that the residual pathway is not mere optimization plumbing; it is part of the model's representational machinery. We agree, but argue that the cleanest…
Higher-order Linear Attention
Yifan Zhang, Zhen Qin, Mengdi Wang +1
The quadratic cost of scaled dot-product attention is a central obstacle to scaling autoregressive language models to long contexts. Linear-time attention and State Space Models (S…
FlashSampling: Fast and Memory-Efficient Exact Sampling
Tomas Ruiz, Zhen Qin, Yifan Zhang +3
Sampling from a categorical distribution is mathematically simple, but in large-vocabulary decoding, it often triggers extra memory traffic and extra kernels after the LM head. We…