# juntang-zhuang/Adabelief-Optimizer

Repository for NeurIPS 2020 Spotlight  "AdaBelief Optimizer: Adapting stepsizes by the belief in observed gradients"

Repository: https://github.com/juntang-zhuang/Adabelief-Optimizer
Canonical: https://ross.abutalabs.com/products/adabelief-optimizer
Language: Jupyter Notebook
License: BSD-2-Clause
License Family: permissive
Last push: 2024-08-09T09:53:49+00:00

## Health v2 (maintenance only)
Score: 32/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 100
- inputs: {"age_days": 2154, "days_push": 754, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1070, forks 107 (observed 2026-08-28T04:03:27.939372+00:00)

## What it is
AdaBelief is a deep learning optimizer that adapts step sizes based on the 'belief' in observed gradients, combining Adam's fast convergence with SGD's generalization. It provides pip-installable packages for both PyTorch (adabelief-pytorch) and TensorFlow (adabelief-tf), published as a NeurIPS 2020 Spotlight paper.

## Use cases
- train neural networks with an optimizer that generalizes like SGD
- find a fast-converging optimizer for GAN training
- replace Adam in image classification experiments
- train transformers with a stable adaptive optimizer
- use a drop-in optimizer for reinforcement learning agents
- benchmark alternative optimizers against Adam and SGD

## When to choose
- you want Adam-like speed with better generalization in PyTorch or TensorFlow
- you are training GANs where Adam is unstable
- you need a well-cited, research-backed optimizer with reference hyperparameters

## When to avoid
- you need a heavily maintained production library with frequent updates
- your framework is not PyTorch or TensorFlow
- you require optimizers beyond AdaBelief in one package

## Facets
- artifact type: library
- maturity: stable
- function: machine-learning, deep-learning
- domain: machine-learning, deep-learning, artificial-intelligence
- platform: python, cross-platform
- tags: optimizer, pytorch, tensorflow, adabelief, gradient-descent, neurips-2020, research

## Member repositories
- juntang-zhuang/Adabelief-Optimizer (main) score 32

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:27.939372+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 2026-08-30T06:54:02.493640+00:00, confidence not recorded.
  - readme: https://github.com/juntang-zhuang/Adabelief-Optimizer (fetched 2026-08-28T04:03:27.939372+00:00, sha c9e1a844ab18)
- Data as of 2026-08-30T08:39:29.467469+00:00.