Building Machine Learning Applications using Hugging Face
summary
Hugging Face shows concrete plumbing: the Hub hosts over 500,000 models and 100,000 datasets and handles scale like a 141 billion-parameter model pushed in under 6 hours. The workflow uses datasets library (up to 15 trillion tokens), TRL for preference-aligned chat fine-tuning, parameter-efficient updates that are a few megabytes so you can fine-tune on a laptop, and a pipeline that produced a new model that was briefly state-of-the-art for half a day.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 2:28Software to Transformers
- 5:56Hugging Face’s ML mission
- 7:39Loading datasets/models
- 9:05Fine-tuning with TRL
- 10:51Demos with Gradio/Space
- 12:34Hub as collaborative platform
- 13:36Open-source in action
- 18:30Q&A
related talks
33mMachine Learning 2.0 with Hugging Face
34mExploring Hugging Face: A Hub for Machine Learning Models
31mHugging Face ecosystem for Local AI/ ML
01h 09mReinventing Machine Learning with Transformers and Hugging Face
49mMachine Learning 2.0 With Hugging Face
14mDiving in Hugging Face Hub; Share Your Model Weights on the #1 AI Hub, Home of 700k+...
32mAn Introduction to Transfer Learning in NLP and HuggingFace
28mUnlock the Power of LLM: Build a HuggingFace Agent