AutoMegaKernel: Compile an LLM into one provably-correct CUDA megakernelgithub.com·4 pts·OsamaJaber·0
StreamIndex: Memory-bounded compressed sparse attention via streaming top-karxiv.org·4 pts·OsamaJaber·0
AutoKernel: Autonomous GPU Kernel Optimization via Iterative Agent-Driven Searcharxiv.org·4 pts·OsamaJaber·0
Tide: Token-Informed Depth Execution for Per-Token Early Exit in LLM Inferencearxiv.org·3 pts·OsamaJaber·1