Disaggregated Quantization: Specializing LLM Prefill and Decode Paper • 2609.26333 • Published 7 days ago • 45
GenFirst: Generation Before Reconstruction for Stable End-to-End Latent Generative Modeling Paper • 2608.29335 • Published Aug 29 • 73
Running 31 Weight-Space Geometry of Offline Reasoning Training 🧠31 Interactive weight-space geometry of six reasoning losses