TensorFold logo

Welcome to TensorFold

TensorFold is an open-source inference engine that runs models on your own hardware and serves them through an OpenAI-compatible API. On Hugging Face, we publish model conversions for Apple silicon, along with previews for new model architectures.

Setup guide · Browse all models · Engine on GitHub

Find a model for your Mac

Start with the collection for your Mac’s unified memory.

New to MLX? Try Nemotron 3.5 Lightning 4-bit. Its card includes a quick start and the conditions for a measured Apple silicon run. Check the runtime and memory notes before downloading. Longer contexts and concurrent requests can use more memory.

What’s new

Tencent Hy4-flash with DFly2 is available as a checkpoint preview in MLX 4-bit and MLX 8-bit. Conversion checks have passed, and TensorFold runtime support is pending.

Need a hand?

Use the Community tab on a model’s page for questions about that checkpoint. For engine issues, use GitHub Issues. Include your hardware and memory, runtime version, and model ID so we can help.

The original model authors receive credit in the checkpoint cards, alongside links to the source models and their licences.