DeepSpec: a full-stack codebase for training and evaluating speculative decoding algorithms

W8A8/W4A8 inference + optimized SDPA on Apple Silicon — unlocking unused INT8 TensorOps in M5 for 1.2–1.9× faster LLM prefill, plus FlashInfer-inspired GQA decode attention for up to 1.6× SDPA spee…

Rust-embedded DSL for writing Apple Metal GPU kernels.

Spiking Neural Network library built natively on Apple MLX

Yet another sparse attention

Alleviating Forgetfulness of Linear Attention by Hybrid Sparse Attention and Contextualized Learnable Token Eviction.

Training neural networks on Apple Neural Engine via reverse-engineered private APIs

Your own personal AI assistant. Any OS. Any Platform. The lobster way. 🦞

Setup guide for ML training on NVIDIA DGX Spark (GB10 Blackwell, CUDA 13, aarch64)

A listing of compiler, language and runtime teams for people looking for jobs in this area

A Swift library for creating and exporting CoreML Models in Swift

Convert StableHLO models into Apple Core ML format

MoE training for Me and You and maybe other people

WIP: Experiments with 2-bit QAT and LoRA

experiments with MLX

a mini 2x2 systolic array and PE demo

🤗 Transformers: the model-definition framework for state-of-the-art machine learning models in text, vision, audio, and multimodal models, for both inference and training.

NanoGPT (124M) in 90 seconds

Test code for my custom float multiply blog post: https://probablydance.com/2025/02/08/why-does-integer-addition-approximate-float-multiplication/

MLX: An array framework for Apple silicon

A machine learning accelerator core designed for energy-efficient AI at the edge.

Hierarchical Reasoning Model Official Release

MLX: An array framework for Apple silicon

Easily fine-tune, evaluate and deploy Qwen, Gemma, or any open weight LLM!

NVIDIA Linux open GPU with P2P support

An open-source coding agent for the Grok API