Supporting PyTorch models with the Google AI Edge TFLite runtime.

LiteRT is the new name for TensorFlow Lite (TFLite). While the name is new, it's still the same trusted, high-performance runtime for on-device AI, now with an expanded vision.

Representation and Reference Lowering of ONNX Models in MLIR Compiler Infrastructure

🚀🚀 「大模型」2小时完全从0训练26M的小参数GPT！🌏 Train a 26M-parameter GPT from scratch in just 2h!

ONNX Script enables developers to naturally author ONNX functions and models using a subset of Python.

A minimal GPU design in Verilog to learn how GPUs work from the ground up

ModernBERT model optimized for Apple Neural Engine.

General-purpose programming language and toolchain for maintaining robust, optimal, and reusable software.

Yet Another Language Model: LLM inference in C++/CUDA, no libraries except for I/O

A Simple Virtual Computer

CUDA/Metal accelerated language model inference

JSON for Classic C++

Book in Progress

A playbook for systematically maximizing the performance of deep learning models.

Modern C++ Programming Course (C++03/11/14/17/20/23/26)

Making the community's best AI chat models available to everyone.

Making C Look ✨Pretty✨and Lua/Lisp/Python-esque

[ICLR 2025] VILA-U: a Unified Foundation Model Integrating Visual Understanding and Generation

Fast and accurate automatic speech recognition (ASR) for edge devices

SANA: Efficient High-Resolution Image Synthesis with Linear Diffusion Transformer

Meta Lingua: a lean, efficient, and easy-to-hack codebase to research LLMs.

Official inference framework for 1-bit LLMs

The official implementation of paper: SimLayerKV: A Simple Framework for Layer-Level KV Cache Reduction.

Swift implementation of Flux.1 using mlx-swift

HART: Efficient Visual Generation with Hybrid Autoregressive Transformer

Modern, extensible Python project management

llama3.np is a pure NumPy implementation for Llama 3 model.

llama3.cuda is a pure C/CUDA implementation for Llama 3 model.