Hugging Face Transformers
Reference implementations and pretrained weights for transformer models, behind one API.
Overview
Transformers is a library providing consistent implementations and pretrained weights for a very large number of model architectures. It solves the problem that every paper ships its own code, so using two models meant learning two codebases. It exists because a shared interface over weights turned model comparison into a few lines, and it is the reason pretrained models became something you download rather than something you reimplement.
Technology Graph
How Hugging Face Transformers connects to the rest of the stack.
Uses and depends on
- Transformer architecture Algorithms Hugging Face Transformers uses Transformer High ✓ source
Tools
Software that implements or integrates with Hugging Face Transformers.