An Introduction to Transfer Learning and Hugging Face
summary
Recent NLP breakthroughs are tied to pretraining on large corpora with Transformer architectures and then adapting them; Hugging Face supplies the actual tools: transformers, tokenizers, datasets and distilled/pruned models. Demonstrations show a 2,500-example MRPC fine-tune hitting >90% accuracy after one epoch and 3.6% error after three, Optuna hyperparameter search for reproducible tuning, Rust tokenizers that process ~1 GB in tens of seconds, and Apache Arrow backed datasets with zero‑copy mapping for huge corpora. Models scale to 175B+ parameters, raising compute and brittleness concerns.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 0:35Transfer learning basics
- 4:20Adaptation and pre-training
- 6:54Transformer architecture
- 11:09Implementation examples
- 18:12Trends and limits
- 24:44Hugging Face ecosystem
- 31:44Fine-tuning tutorial
- 40:46Q&A session
related talks
32mAn Introduction to Transfer Learning in NLP and HuggingFace
01h 09mReinventing Machine Learning with Transformers and Hugging Face
02h 10mIntro to Hugging Face: Fine-tuning BERT for NLP tasks
45mTransfer Learning - Entering a new era in NLP
27mEmbeddings, Transfer Learning and More: LLM Tailored to You
49mMachine Learning 2.0 With Hugging Face
31mHugging Face ecosystem for Local AI/ ML
34mTransformers End to End Experience training...