4 papers
ADPO: Anchored Direct Preference Optimization
Wang Zixian
We present Anchored Direct Preference Optimization (ADPO), a policy alignment method derived from first principles of KL-regularized reinforcement learning. Unlike standard approac…
APO: Alpha-Divergence Preference Optimization
Wang Zixian
Two divergence regimes dominate modern alignment practice. Supervised fine-tuning and many distillation-style objectives implicitly minimize the forward KL divergence KL(q || pi_th…
Attention Saturation and Gradient Suppression at Inflection Layers: Diagnosing and Mitigating Bottlenecks in Transformer Adaptation
Wang Zixian
Pre-trained Transformers often exhibit over-confidence in source patterns and difficulty in forming new target-domain patterns during fine-tuning. We formalize the mechanism of out…
Omniwise: Predicting GPU Kernels Performance with LLMs
Zixian Wang, Cole Ramos, Muhammad A. Awad +1
In recent years, the rapid advancement of deep neural networks (DNNs) has revolutionized artificial intelligence, enabling models with unprecedented capabilities in understanding,…