Silicon Innovation Day 2023 | AWS On Air ft. AWS Silicon Hugging Face collaboration
summary
Collaboration ports Hugging Face libraries to AWS Trainium and Inferentia, using Optimum Neuron and Deep Learning Containers to run real jobs not demos. Measured gains include 5x training throughput on Trainium 1 versus comparable GPU, 8x inference on Inferentia 2 for BERT and 2x for ViT, and a BERT fine‑tune that finishes in 9 minutes for $0.20 on a trn1. Instances are available via the Hugging Face AMI in AWS Marketplace and integrate with SageMaker and the Hugging Face Hub of ~250k models.
AI-written summary · about
related talks
25mSponsored Session: Accelerating GenAI Inference: From AWS Deep Learning...
49mMachine Learning 2.0 With Hugging Face
54mDeep learning on AWS with NVIDIA: From training to deployment
42mFine-tune and deploy LLMs from Hugging Face (AIM304)
33mMachine Learning 2.0 with Hugging Face
14mKeynote: Building AI with Open Source & Hugging Face- 14mKeynote: Building AI with Open Source & Hugging Face
31mRealizing Industrial AI at Scale with