Repository brief

jingyaogong/minimind

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 2026-03-30T12:08:39.157Z
3mo ago

jingyaogong/minimind

MiniMind is a highly popular Python open source project for training a very small GPT-style language model from scratch. The repository emphasizes a full LLM training stack rather than just inference, with code and data around pretraining, SFT, LoRA, RLHF/RLAIF, tool use, agentic RL, distillation, evaluation, and a minimal OpenAI-compatible server and chat UI. It is actively maintained, not archived, and has strong adoption signals with 44,715 stars and 5,389 forks.

GitHub
Loading tags...
Stars44,715
Forks5,389
Default branchmaster
Last pushed2026-03-27T13:20:12Z
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

Choose this fork only if you want a lightweight, mostly static MiniMind snapshot and are fine giving up upstream freshness. For anyone planning active training, evaluation, or serving work, upstream is the better default.

Choose this fork if you want a smaller, experiment-friendly MiniMind branch with extra sample and function-calling material. Choose upstream if you want the fuller, actively maintained training stack and the newest fixes.

Choose this fork if your priority is comprehension and walkthroughs; choose upstream if you want the most current, broadly maintained training stack.

Choose this fork if your priority is the smallest practical MiniMind variant and you do not need the latest upstream fixes. Stick with upstream if you want the most current training stack, defaults, and maintenance.

Prefer upstream unless you specifically want this fork’s 26M positioning or a frozen snapshot; otherwise it is mostly a lagging copy without demonstrated functional gains.

Prefer upstream if you want the current, complete MiniMind stack. Prefer this fork only if you specifically want a smaller 26M-oriented training record and do not need the latest upstream functionality.

Prefer upstream unless you specifically need this older snapshot; this fork adds no visible features and is materially behind recent upstream work.

Adopt only if you specifically want a smaller, older MiniMind snapshot and do not need upstream’s latest fixes or features. For most users, upstream is the safer choice.

Prefer this fork only if you care about the readability-oriented `einops` refactor and want a near-upstream codebase. If you need the newest upstream fixes or the broadest feature set, upstream is the safer choice.

Choose this fork only if you specifically want the smaller 26M framing and are comfortable with a stale, near-upstream snapshot. Otherwise, upstream is the safer choice because it is more current and materially richer in maintained features.

jingyaogong/minimind · Discofork