Ruisi Cai

$\nabla$-Reasoner: LLM Reasoning via Test-Time Gradient Descent in Latent Space

$Read-ME$: Refactorizing LLMs as Router-Decoupled Mixture of Experts with System Co-Design