Lists (3)
Starred repositories
reproducing the flapping airplanes megakernel, in collab with magni
🧱 Experimental library for building inference engines
Sub-1-bit LLM compression (LittleBit / LittleBit-2) re-implemented in PyTorch, with an attention-sink-aware distillation loss. Qwen3-0.6B at 0.55 BPW: PPL 36.4 vs 21.0 on a GTX 1080.
High-performance safetensors model loader
A PyTorch native automated mega-kernel compiler
Run the full 2.78-trillion-parameter Kimi K3 model, DeepSeek V4.1 Flash or GLM-5.3-Flash beyond available RAM by streaming activated weights directly from NVMe. A dependency-free, embeddable C infe…
Development repo for translating Software Foundations to Lean
Optimized GPU compiler for LLM inference. Choose from a list of optimized recipes or optimize your own model via kernel fusion, autotuning, and advanced scheduling. Run benchmarks across different …
Autoresearch for GPU kernels. Give it any PyTorch model, go to sleep, wake up to optimized Triton kernels.
A local inference engine for Apple silicon, built around the model.
Mixture-of-experts (MoE) training megakernel for NVL72s
A curated list of Lean 4 libraries, tools, frameworks and learning resources for programmers
A high-throughput and memory-efficient inference and serving engine for LLMs
An independent implementation of the Lean 4 kernel on .NET, and a checker for Lean exports
High performance coding agent harness written in Rust, built for heavy multi-agent dev workflows
A Rust framework for correct and performant distributed systems
A unified CLI to connect artificial intelligence to the social web.
why-did-you-render by Welldone Software monkey patches React to notify you about potentially avoidable re-renders. (Works with React Native as well.)
Shen extension that makes layout overflow a compile-time type error
Halo is an open-source framework built by White Circle for training large language and multimodal models