(Neural) Natural Language Processing in Go
summary
A self-contained NLP library written in pure Go that can both train and run neural models without Python or cgo, plus a model importer that converts Hugging Face checkpoints into spaGO format. Demoed features: NER with ~93–94% accuracy, span-based question answering with per-answer confidence scores (examples at 0.93–0.96), zero-shot text classification, cross-lingual similarity and machine translation in ~22 lines. Models load as a server (models commonly 1–2 GB), or deploy as a 20 MB docker image for tiny runtime footprints; examples live in the GitHub gophercon repo.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:45What is Pago
- 4:58Gradients
- 6:13Neural Models
- 7:39Use Cases
- 9:08Python
- 10:33Neural Model Importer
- 14:36Language Model
- 16:51Must Language Model
- 18:38Question Answering
- 21:12Text Classification
- 23:32Text Similarity
- 25:23Machine Translation
- 26:27Conclusions Future Words
related talks
33mspaGO: a work-in-progress yet ready-to-use library for ML/NLP
29mThe Go-to Language for AI: Opportunities and Challenges
50mDeep Learning with Go
04mGo and data, can they ever be friends?
20mUsing TensorFlow with Go
31mMachine Learning for Gophers
29mLearn Neural Networks With Go, Not Math
29mThe Gopher Orchestra: Distributed Machine Learning with Go