Intel's profile picture

Intel

company
Verified

AI & ML interests

None defined yet.

Organization Card
About org cards

Intel and Hugging Face are building powerful optimization tools to accelerate training and inference with Transformers.

Learn more about Hugging Face collaboration with Intel AI
Quantize Transformers with Intel® Neural Compressor and Optimum
Quantizing 7B LLM on Intel CPU

Intel optimizes widely adopted and innovative AI software tools, frameworks, and libraries for Intel® architecture. Whether you are computing locally or deploying AI applications on a massive scale, your organization can achieve peak performance with AI software optimized for Intel® Xeon® Scalable platforms.

Intel’s engineering collaboration with Hugging Face offers state-of-the-art hardware and software acceleration to train, fine-tune and predict with Transformers.

Useful Resources:

 

Get Started

1. Intel Acceleration Libraries

To get started with Intel hardware and software optimizations, download and install the Optimum Intel and Intel® Extension for Transformers libraries. Follow these documents to learn how to install and use these libraries:

The Optimum Intel library provides primarily hardware acceleration, while the Intel® Extension for Transformers is focused more on software accleration. Both should be present to achieve ideal performance and productivity gains in transfer learning and fine-tuning with Hugging Face.

2. Find Your Model

Next, find your desired model (and dataset) by using the search box at the top-left of Hugging Face’s website. Add “intel” to your search to narrow your search to models pretrained by Intel.

3. Read Through the Demo, Dataset, and Quick-Start Commands

On the model’s page (called a “Model Card”) you will find description and usage information, an embedded inferencing demo, and the associated dataset. In the upper-right of your screen, click “Use in Transformers” for helpful code hints on how to import the model to your own workspace with an established Hugging Face pipeline and tokenizer.