# jingyaogong/minimind-o

🎙️ A 0.1B Omni model trained from scratch, capable of listening, speaking, and seeing!

Repository: https://github.com/jingyaogong/minimind-o
Canonical: https://ross.abutalabs.com/products/minimind-o
Homepage: https://jingyaogong.github.io/minimind-o
Language: Python
License: Apache-2.0
License Family: permissive
Topics: artificial-intelligence, chatgpt, omni
Last push: 2026-08-06T14:08:34+00:00

## Health v2 (maintenance only)
Score: 57/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 96, release rhythm 35, longevity 8
- inputs: {"age_days": 124, "days_push": 27, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, young
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2384, forks 280 (observed 2026-08-28T04:06:42.440238+00:00)

## What it is
MiniMind-O is an open-source educational project that implements a tiny (~0.1B parameter) end-to-end Omni model from scratch in PyTorch, handling text, image, and audio inputs with text and streaming speech outputs. It includes full training code, model weights, mini/full datasets, and a technical report, designed so a single RTX 3090 can run the complete training pipeline in about 2 hours.

## Use cases
- train a small multimodal omni model from scratch
- learn how thinker-talker speech architectures work
- build a model that listens, sees, and speaks
- experiment with streaming voice generation and barge-in interruption
- study voice cloning with reference audio codes
- run a lightweight GPT-4o-style model on a personal GPU
- understand end-to-end speech-text hidden state fusion

## When to choose
- you want to read, modify, and train a complete omni model from first principles
- you have a single consumer GPU and limited time/budget
- you need a readable baseline for multimodal speech research or teaching
- you want full open access to code, weights, data, and a technical report

## When to avoid
- you need production-grade speech quality or low-latency real-time deployment
- you want a plug-and-play inference library rather than a training/learning codebase
- you need large-scale model capacity or broad multilingual coverage
- you require enterprise support or long-term stability guarantees

## Facets
- artifact type: learning-resource
- maturity: active
- function: machine-learning, deep-learning, llm-training, speech-recognition, tts, chatbot, audio-processing, computer-vision
- domain: artificial-intelligence, large-language-models, speech-processing, computer-vision, education, tutorials
- platform: python, cross-platform
- tags: omni-model, multimodal, thinker-talker, streaming-voice, voice-cloning, train-from-scratch, pytorch, mtp, small-language-model, natural-language-processing, gpu

## Member repositories
- jingyaogong/minimind-o (main) score 57

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:42.440238+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 2026-08-30T02:34:39.952596+00:00, confidence not recorded.
  - readme: https://github.com/jingyaogong/minimind-o (fetched 2026-08-28T04:06:42.440238+00:00, sha 2a77c6efbcbd)
  - homepage: https://jingyaogong.github.io/minimind-o (fetched 2026-08-29T10:15:31.770800+00:00, sha 2cf794458036)
- Data as of 2026-08-30T08:39:29.467469+00:00.