Latent Feedback Is a Compute and Data Amplifierwith Benhao Huang, Chaitanya Dwivedi, Xuezhe Ma, Hector Liu1 versionW_wu7dhj2yv1 · current
Taming Softmax at the Limits of Muon: A Unified Approach through Hilbert GeometryAditya Somasundaram, Charles Guille-Escuret, Alexander Moreno et al.0 citations2610.17588v12026arXiv
Unlocking Lossless Speedups in LLMs via Discrete DiffusionSubham Sekhar Sahoo, Lingjie Chen, Khiem Pham et al.0 citations2609.04010v12026arXiv
Toward a First-Principles Update Geometry for the Language-Model HeadAditya Somasundaram, Charles Guille-Escuret, Alexander Moreno et al.0 citations2608.22253v22026arXiv
GQA-{\mu}P: The maximal parameterization update for grouped query attentionKyle R. Chickering, Huijuan Wang, Mengxi Wu et al.0 citations2605.15290v12026arXiv