Machine Learning 2.0 With Hugging Face
summary
A concrete workflow shows up: DistilBERT fine-tuned on 100k Amazon shoe reviews (20k per star), trained for a quick epoch on GPU, pushed to the Hugging Face Hub and served via push-to-hub + Inference API, then wrapped in a 113-line Gradio Space demo. Latency tuning uses optimum + ONNX quantization, shrinking a model from ~256MB to ~165MB and cutting inference time (500 predictions from ~14s to much less) while keeping accuracy within a 3% loss budget.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:06Deep learning to zero
- 4:38What is ML 2.0
- 7:29Transformers and Hugging Face
- 10:59Agile ML workflow
- 13:49Hardware acceleration
- 15:55Demo setup
- 19:23Amazon reviews dataset
- 22:56Fine-tuning DistilBERT
- 27:51Model evaluation and deployment
- 31:44Evaluate library
- 32:47Hugging Face Spaces demos
- 39:11Inference performance
- 46:12Next steps
related talks
33mMachine Learning 2.0 with Hugging Face
31mHugging Face ecosystem for Local AI/ ML
19mBuilding Machine Learning Applications using Hugging Face
01h 09mReinventing Machine Learning with Transformers and Hugging Face
02h 10mIntro to Hugging Face: Fine-tuning BERT for NLP tasks
21mFrom 0 to a crowd sourced neural network in a jiffy
58mAn Introduction to Transfer Learning and Hugging Face
05mSilicon Innovation Day 2023 | AWS On Air ft. AWS Silicon Hugging Face collaboration