Deploy Hugging Face Models on Azure powered by AMD ROCm software
summary
Azure ND MI300X v5 VMs with AMD ROCm run Hugging Face code unchanged: copy-paste from Hugging Face loads ResNet (small), a 600M Bart summarizer, Whisper ASR, a 1.5B speech model, Phi instruct models, SDxcel Turbo text-to-image and an ~11B text-to-video model, and even pretrains GPT-2 on WikiText on a single GPU. Frameworks like PyTorch/TensorFlow/JAX, vLLM and ONNX work out of the box with no porting or recompilation. One MI300X GPU can load up to 70 billion parameters and eight such GPUs sit in the instance.
AI-written summary · about
chapters (10)
- 0:00Introduction to AMD and AI
- 1:30Image classification demo
- 2:39NLP summarization demo
- 4:32Translation capabilities
- 5:21Automatic speech recognition
- 6:41Small language models
- 8:08Text-to-image generation
- 9:28Text-to-video demo
- 10:51Model training demonstration
- 12:10Hardware scale and summary
related talks
31mHugging Face ecosystem for Local AI/ ML
32mPowering Your Generative AI Workloads with AMD and Open-Source ROCm
28mHarnessing generative AI with NVIDIA AI and Microsoft Azure
13mFrom Silicon to Scale: Training at Scale
22mAI @Scale
19mBuilding Machine Learning Applications using Hugging Face
05mBuild AI Anywhere with ROCm™ Software: AMD and PyTorch Bring...
12mLightning Talk: PyTorch 2.0 on the ROCm Platform