# explosion/thinc

🔮 A refreshing functional take on deep learning, compatible with your favorite libraries

Repository: https://github.com/explosion/thinc
Canonical: https://ross.abutalabs.com/products/thinc
Homepage: https://thinc.ai
Language: Python
License: MIT
License Family: permissive
Topics: machine-learning, deep-learning, artificial-intelligence, ai, python, nlp, natural-language-processing, spacy, machine-learning-library, type-checking, functional-programming, pytorch, tensorflow, mxnet, jax
Last push: 2026-03-27T08:50:19+00:00

## Health v2 (maintenance only)
Score: 80/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 74, release rhythm 76, longevity 100
- inputs: {"age_days": 4339, "days_push": 159, "days_rel": 163, "gap_med": 17, "n_releases_24m": 12}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2890, forks 294 (observed 2026-08-28T04:07:28.358177+00:00)

## What it is
Thinc is a lightweight deep learning library from the makers of spaCy offering a type-checked, functional-programming API for composing neural network models. It wraps PyTorch, TensorFlow, and MXNet layers so you can mix frameworks in a single model and configure model trees declaratively.

## Use cases
- compose neural network models with a functional API in Python
- wrap PyTorch or TensorFlow models inside another pipeline
- build spaCy components using my preferred ML framework
- type-check model definitions with mypy
- configure model architectures and hyperparameters declaratively
- train custom NLP models without a heavyweight framework

## When to choose
- you need a lightweight interface layer between frameworks like PyTorch and TensorFlow
- you're developing spaCy or Prodigy components
- you want type-checked, functional model composition
- you want a clean config system for ML object trees

## When to avoid
- you need a full standalone deep learning framework with a large ecosystem
- you rely on cutting-edge model architectures only available in PyTorch or JAX
- your team prefers imperative, class-based model definitions

## Facets
- artifact type: library
- maturity: active
- function: deep-learning, machine-learning, nlp, type-system, configuration-management
- domain: deep-learning, machine-learning, artificial-intelligence
- platform: python, windows
- tags: functional-programming, model-composition, pytorch-wrapper, tensorflow-wrapper, mxnet, jax, spacy, mypy-plugin, config-system, natural-language-processing, linux, macos, gpu

## Member repositories
- explosion/thinc (main) score 80

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:07:28.358177+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T07:35:20.175452+00:00, confidence not recorded.
  - readme: https://github.com/explosion/thinc (fetched 2026-08-28T04:07:28.358177+00:00, sha 95d453e5aa77)
  - homepage: https://thinc.ai (fetched 2026-08-29T09:50:34.288449+00:00, sha 170fd77392ad)
  - site_page: https://thinc.ai/docs (fetched 2026-08-29T09:50:34.297298+00:00, sha c5f773ac3450)
  - site_page: https://thinc.ai/docs/usage-frameworks (fetched 2026-08-29T09:50:34.299340+00:00, sha 4f31065bb1a0)
- Data as of 2026-08-30T08:39:29.467469+00:00.