ML engineers deploying Hugging Face models who need faster or smaller inference on specific hardware.

A Hugging Face toolkit that exports, quantizes, and accelerates Transformers, Diffusers, and Sentence Transformers models for ONNX Runtime, OpenVINO, TensorRT, and specialized chips.

Used in Python pipelines targeting CPUs, GPUs, Intel, AWS Inferentia, and other accelerators.

Use when a trained model needs to be compressed or exported before production deployment.

It closes the gap between a research checkpoint and an efficient production inference target.

Built on top of Hugging Face Transformers and integrates with ONNX Runtime, OpenVINO, and vendor SDKs.

Open source, Apache 2.0 licensed, maintained by Hugging Face; no credentials required for local use.