# facebookresearch/higher

higher is a pytorch library allowing users to obtain higher order gradients over losses spanning training loops rather than individual training steps.

Repository: https://github.com/facebookresearch/higher
Canonical: https://ross.abutalabs.com/products/higher
Language: Python
License: Apache-2.0
License Family: permissive
Archived: true
Last push: 2022-03-25T15:56:51+00:00

## Health v2 (maintenance only)
Score: 10/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 100
- inputs: {"age_days": 2553, "days_push": 1622, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, archived
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1626, forks 125 (observed 2026-08-28T04:05:13.074409+00:00)

## What it is
higher is a PyTorch library that enables higher-order gradients over losses spanning entire training loops rather than individual steps. It provides tools to make torch.nn modules stateless and differentiable versions of torch.optim optimizers for meta-learning approaches.

## Use cases
- implement meta-learning algorithms like MAML in pytorch
- compute gradients through unrolled optimization loops
- differentiate through adam or sgd update steps
- make torch nn modules functional for gradient tracking
- hyperparameter optimization via gradient-based methods
- second-order gradient computation across training steps

## When to choose
- you need to backpropagate through multiple model update steps
- you are implementing meta-learning or gradient-based hyperparameter optimization
- you want differentiable optimizers compatible with existing torch code

## When to avoid
- you only need standard first-order training without gradients through updates
- you are not using pytorch
- you need actively maintained tooling, as the last release was in 2022

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, deep-learning
- domain: machine-learning, deep-learning, artificial-intelligence
- platform: python
- tags: pytorch, meta-learning, differentiable-optimization, higher-order-gradients, hyperparameter-optimization

## Member repositories
- facebookresearch/higher (main) score 10

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:13.074409+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 2026-08-30T03:48:27.085537+00:00, confidence not recorded.
  - readme: https://github.com/facebookresearch/higher (fetched 2026-08-28T04:05:13.074409+00:00, sha cee0bd826cab)
  - registry_pypi: https://pypi.org/pypi/higher/json (fetched 2026-08-29T11:21:14.643872+00:00, sha 97cd73f74268)
- Data as of 2026-08-30T08:39:29.467469+00:00.