Stars
Wake and Break: a PSP exploit for AMD navi31 (RX 7900) that makes the card accept unsigned IP firmware. Not persistent: a cold power cycle restores stock.
DeepInfer: High-performance, large-scale inference framework for DeepSeek models
PokéBench: local and frontier models play Pokémon Red as a benchmark, with a leaderboard and live stream
A hierarchical inference engine for frontier MoE models: hot experts on the GPU (CUDA, Vulkan), warm ones in RAM, cold ones streamed from NVMe. Written in C, runs on the hardware you already own.
slime is an LLM training framework for RL Scaling.
KernelBench: Can LLMs Write GPU Kernels? - Benchmark + Toolkit with Torch -> CUDA (+ more DSLs)
A single CLAUDE.md file to improve Claude Code behavior, derived from Andrej Karpathy's observations on LLM coding pitfalls.
A kernel library written in tilelang
OpenCode 2 plugin that uses your existing Claude Code credentials — no separate login needed.
Open-source CUDA, Triton and HIP compiler targeting multiple GPU and CPU architectures.
Documentation for the Mainboard and printable mechanical parts in the Framework Desktop
A project trying to build a hoverboard controller without semiconductors
Code, labs, and resources for O'Reilly AI Systems Performance Engineering: GPU optimization, distributed training, inference scaling, and full-stack tuning.
A machine learning accelerator core designed for energy-efficient AI at the edge.
Memory Optimizations for Deep Learning (ICML 2023)
Exocompilation for productive programming of hardware accelerators
Mirage Persistent Kernel: Compiling LLMs into a MegaKernel
Minimal reproduction of DeepSeek R1-Zero
Open-source high-performance RISC-V processor
Type annotations and runtime checking for shape and dtype of JAX/NumPy/PyTorch/etc. arrays. https://docs.kidger.site/jaxtyping/
the official Rust and C implementations of the BLAKE3 cryptographic hash function