spago

Why we included this project

Most of the deep-learning ecosystem assumes you will orchestrate a Python runtime, so Spago is a useful exception: a self-contained Go library with automatic differentiation over a define-by-run computation graph, plus the usual building blocks like LSTM, GRU, and attention layers. Because everything compiles into a single Go binary, inference embeds cleanly into services and CLIs that run on machines where installing Python or heavy dependencies is impractical. The codebase favors clarity over raw speed, which makes it a readable reference for how transformers and gradient computation actually work. Teams already on Go can also start from built-in models for classic NLP tasks such as named-entity recognition, translation, and question answering.