/jump to repo

Repos Recent Bookmarks Watched Notes Tags Discover Compare Stats GitHub

Repository brief

tensorflow/models

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 2026-03-30T12:13:51.628Z

3mo ago

tensorflow/models

tensorflow/models is TensorFlow’s Model Garden: a large, actively updated repository of model implementations and examples. It includes officially maintained TensorFlow 2 model code in `official`, researcher-maintained implementations in `research`, a curated `community` directory, and `orbit` for custom TensorFlow 2 training loops. It has very high adoption signals, with 77,681 stars and 45,190 forks, and recent commits in March 2026.

Loading tags...

Stars77,681

Forks45,190

Default branchmaster

Last pushed2026-03-25T22:31:46Z

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

Maintenance:

Magnitude:

Sort:

Selected

Choose this fork only if you specifically need its preserved research data and project artifacts. For most adopters, upstream is the better default because this fork is stale, materially divergent, and likely missing recent fixes and API modernization.

Prefer this fork if your goal is music object detection and you want a pre-shaped project with configs, demos, and inference workflow already centered on that use case. Prefer upstream if you need current TensorFlow Model Garden coverage, ongoing maintenance, and broader reusable model examples.

Prefer this fork only if you specifically need a frozen legacy snapshot with bundled historical NLP/vision assets. If you want an actively maintained TensorFlow Model Garden, upstream is the better choice.

Prefer this fork only if you need its legacy custom NLP/data-serving workflow and are willing to own a stale, highly divergent codebase. For new work, upstream is the better default because it is active, broader, and much more current.

Prefer this fork only if you need the legacy SyntaxNet/DRAGNN content or a frozen 2017 research snapshot. If you want a current, supported TensorFlow models repo, upstream is the better choice by a wide margin.

Prefer this fork only if you need the older legacy code and datasets exactly as they were in 2017. If you want a maintained TensorFlow Model Garden, current TF2 examples, or active upstream support, the upstream repository is the better choice.

Choose this fork only if you specifically need the adaptive-sampling segmentation work. For general TensorFlow model development, upstream is far better supported, much broader, and much more current.

Choose this fork only if you need legacy TensorFlow model code or archived experiment assets. For anything current, maintained, or TF2-focused, upstream tensorflow/models is the better choice.

Choose this fork only if you need the historical, legacy TensorFlow Models snapshot. For new work, the upstream repository is the better default because it is active, current, and has modern TensorFlow 2 and packaging support.

Choose this fork only if you specifically need the legacy research additions it preserves. If you want current TensorFlow Model Garden capabilities, active maintenance, or modern TF2 workflows, upstream is the better choice.

Fork comparison

Sarasra/models

38/100

stale

significant_divergence

Choose this fork only if you specifically need its preserved research data and project artifacts. For most adopters, upstream is the better default because this fork is stale, materially divergent, and likely missing recent fixes and API modernization.

Likely purpose

To preserve and distribute a customized TensorFlow models tree with bundled research datasets, model resources, and project-specific docs for users who need the older codebase and offline assets rather than the current upstream release.

Best for

Researchers who need legacy TensorFlow model resources exactly as captured in this fork.; Teams maintaining an older internal workflow built around SyntaxNet/DRAGNN or frozen Inception assets.; Users who value a self-contained snapshot over staying current with upstream TensorFlow APIs.; Adopters comfortable owning a large, divergent ML codebase rather than consuming a maintained upstream release.

Additional features

Bundled SyntaxNet/DRAGNN training and evaluation data for English, Spanish, and Chinese parsing/segmentation workflows.
Included Inception dataset metadata and label resources, which makes image-classification examples easier to run without reconstructing those assets.
Project-specific documentation and website assets under `official/projects/waste_identification_ml`, indicating extra applied-ML workflows beyond the upstream model garden.
Large amounts of checked-in model/resource files that likely support reproducible offline demos or fixed research snapshots.
Expands API, schema, or SDK surface area for downstream integrations and generated clients.
Adds plugin, tool, or MCP-style extensibility beyond the upstream baseline.

Missing features

It is 298 commits behind upstream, so it likely misses newer Model Garden changes, fixes, and maintenance work.
It predates recent upstream maintenance such as the gsutil-to-gcloud storage migration and the pseudo-label processing fix.
It likely lacks current TensorFlow 2 API updates and the latest officially maintained `official` code evolution from upstream.
Newer pip-package and dependency updates reflected in active upstream maintenance are probably absent or harder to use here.
Some upstream behavior appears to be removed, disabled, or intentionally stripped down in this fork.
Some upstream plugin, tool, or MCP-style integration paths may be reduced or disabled.

Strengths

Useful if you need the exact older research assets and parser resources that are no longer convenient to rebuild.
May be better for offline or reproducible workflows because many large data and map files are checked in.
Contains substantial domain-specific additions rather than only cosmetic changes.

Risks

Very stale relative to upstream despite its size, so maintenance burden is high.
Huge divergence makes rebasing or cherry-picking upstream fixes expensive.
Bundled data and generated artifacts can increase repository weight and complicate setup or updates.
Users expecting the current TensorFlow Model Garden experience will likely hit compatibility gaps.