Stories by OsamaJaber

Compiles any HuggingFace model into a single persistent megakernel

Mega Kernels, Written by Agents

AutoMegaKernel: Compiling a LLM into a single CUDA kernel

AutoMegaKernel: Compile an LLM into one provably-correct CUDA megakernel

StreamIndex: Memory-bounded compressed sparse attention via streaming top-k

Show HN: AutoKernel, Auto GPU Kernel Optimization

DeepSeek V4's indexer dies at 65K. We got it to 1M on 6GB

AutoKernel: Autonomous GPU Kernel Optimization via Iterative Agent-Driven Search

DeepSeek V4's indexer OOMs at 65K context. We got it to 1M in 6G

Ouroboros: Dynamic Weight Generation for Recursive Transformers

Tide: Token-Informed Depth Execution for Per-Token Early Exit in LLM Inference

Own your AI. Optimized down to the kernel

Agents with "Hands"

Open-Source Agent Operating System

PicoLM: Run a 1B parameter LLM on a $10 board

The Floating Dock for Developers