Using Hugging Face pre-trained models#
This section shows you how to run pre-trained Hugging Face models. Hugging Face is large repository of pre-trained models, it is also a hub where the AI community collaborates.
Hugging Face models will be stored in:
$USER/.cache/huggingface
%USERPROFILE%\.cache\huggingface\hub
Note
In Docker, the models will be stored in /ROCM_APP/models/hf, this ensures that the model will be only downloaded once, even after stopping the Docker container.
See Hugging Face Environment Variables for more information.
Examples#
Make sure the environment is setup.
This guide provides a number of examples you can run out of the box. Navigate to get-started/hf/ and lunch jupyter lab. Some of of these notebooks can only run if you have an AMD GPU in your system.
- MNIST Classification with an MLP Hugging Face Model
- Semantic Segmentation
- Image Classification using Yolov10
- Fine-tuned LAnguage Net Text-To-Text Transfer Transformer
- Google Enhanced Multimodal Machine Learning 2
- OpenAI Whisper - Speech Recognition
- Phi-3-vision Instruct Open Multimodal Model
- Phi-3 Instruct Open Model
More Pre-trained Models#
If you would like to run a different model, Hugging Face hosts hundred of thousands of models. Explore what Hugging Face has to offer https://huggingface.co/models.
Learn how to download Hugging Face models.
Copyright (C) 2025 Advanced Micro Devices, Inc. All rights reserved.
SPDX-License-Identifier: MIT