Discofork
Repository brief

keras-team/keras

Read the upstream summary on the left, browse the cached forks below it, and load each fork comparison into the right-hand panel.

Cached analysis
cached 2026-03-30T11:56:41.310Z
3mo ago

keras-team/keras

Keras is a large, actively maintained Python deep learning framework. The repo describes Keras 3 as a multi-backend library for JAX, TensorFlow, PyTorch, and OpenVINO, with broad model-building support and a strong focus on high-level usability. It is widely adopted, with 63,917 stars and 19,744 forks, and was updated very recently.

GitHub
Loading tags...
Stars63,917
Forks19,744
Default branchmaster
Last pushed2026-03-29T17:05:31Z
Recommended shortcuts

Jump straight into Discofork's strongest cached fork picks, or open a compare view in one click.

Forks

Choose a fork to inspect

10 of 10 fork briefs
Selected

Prefer this fork only if you need legacy Keras behavior or Caffe/Theano-era code paths. If you want an actively maintained framework, modern backend support, or current ecosystem compatibility, upstream Keras is the better fit.

Choose this fork only if you need legacy Keras compatibility, especially MXNet-era workflows. For new work or active maintenance, upstream Keras is the better default because this fork is stale and far behind on modern backends and ongoing development.

Choose this fork only if you need legacy Theano-era Keras behavior or its historical 3D-convolution work. For new projects or active maintenance, upstream Keras is the better choice by a wide margin.

Choose this fork only if you must preserve legacy Keras behavior. For any new work or active maintenance, upstream Keras is the better default because this fork is far behind and missing the modern multi-backend feature set.

Prefer this fork only if you specifically need legacy Keras behavior. If you want a current deep-learning framework, upstream Keras is the better choice because this fork is far behind and diverged from modern multi-backend Keras.

Choose this fork only if you specifically want its trimmed, custom branch and are prepared to maintain it. For most users, upstream Keras is the better default because this fork is materially stale and has removed major backend/export functionality.

Prefer this fork only if you need to preserve a legacy Theano-era Keras workflow or a fork-specific scientific patch. For new development, upstream Keras is the better choice: it is active, far more feature-complete, and supports modern multi-backend training and inference.

Prefer this fork only if you need old Keras-era behavior or are maintaining a legacy codebase. For new work, upstream Keras is the better choice because this fork is heavily outdated and diverged from the modern multi-backend framework.

Choose this fork only if you need legacy Keras behavior and are willing to live without modern Keras 3 capabilities. For new work, upstream is the clear choice.

Prefer this fork only if you need its Bayesian RNN and MC-dropout-specific behavior for legacy research. For almost any production, backend-portable, or actively maintained Keras use case, upstream is the better choice.