Metal Python Packages

quadrants

High-performance multi-platform compiler for physics simulation

wgpu

WebGPU for Python

veloxquant-mlx

Fast KV-cache quantization for Apple Silicon (MLX) — 43 research-adapted compression methods with Metal kernels

pygfx

Powerful and versatile visualization for Python.

xllamacpp

xllamacpp - a Python wrapper of llama.cpp

metal-linalg

QR, symmetric eigendecomposition and SVD for batches of matrices on Apple Silicon GPUs, for MLX and PyTorch. 2x to 10x gain in performance. Metal kernels per regime, routed to the GPU or the CPU by policies measured per Mac. C++, Python (pip: metal-linalg, metal-linalg-torch), Swift and C.

gstaichi

High-performance multi-platform compiler for physics simulation

mojolearn

GPU machine learning in Mojo that supports bitwise identical inference and training on Apple Metal, NVIDIA CUDA and AMD HIP.

apache-tvm

Open Machine Learning Compiler Framework

mlx-kquant

Native K-quant support for MLX, with a quantization and fine-tuning toolchain for Apple Silicon

lucid-dl

Lumerico's Comprehensive Interface for Deep Learning

pyclesperanto

GPU-accelerated Image Processing library for Python

llama-cpp-py-sync

Auto-synced CFFI ABI python bindings for llama.cpp with prebuilt wheels (CPU/CUDA/Vulkan/Metal).

mtplx

The fastest way to run Qwen 3.8 Flash Next, Qwen 3.8 27B and Ternary Bonsai 2 27B on a Mac: 125 tok/s in OpenCode on an M5 Max, and a 27B model on 16 GB Macs. Native MTP speculative decoding on Apple Silicon, exact at any temperature. OpenAI and Anthropic compatible local server.

demucs-mlx

Split any song into stems — vocals, drums, bass, and more. Fast music source separation for Apple Silicon, powered by MLX.

gmlx

Local inference platform for K/IQ-quant GGUF models on Apple Silicon

mlx-spectro

High-performance STFT/iSTFT for Apple MLX with fused Metal kernels and autograd support

mojolearn-nvidia

GPU machine learning in Mojo that supports bitwise identical inference and training on Apple Metal, NVIDIA CUDA and AMD HIP.

ax-engine

One Mac process. Many models. Real speed. Multi-model LLM serving with prefix reuse, MTP acceleration, and OpenAI APIs — built for Apple Silicon, measured against mlx-lm and llama.cpp.

gafime

GPU-Accelerated Feature Interaction Mining Engine (GAFIME) — native heterogeneous execution across Rust CPU/SIMD, CUDA, ROCm/HIP, and Metal.

mlx-dspark

Up to 4× faster LLM decoding on Apple Silicon, lossless. Native MLX port of DeepSeek's DSpark & z-lab's DFlash speculative decoding — Gemma-4, Qwen3.8, Muse-Glimmer, Nemotron, LFM2.5, Ornith-1.0, ternary Bonsai-27B.

ttnn

:metal: TT-NN operator library, and TT-Metalium low level kernel programming model.

algan-quadrants

High-performance multi-platform compiler for physics simulation

mojolearn-amd

GPU machine learning in Mojo that supports bitwise identical inference and training on Apple Metal, NVIDIA CUDA and AMD HIP.