OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Minimind O

An open-source, 0.1B-scale multimodal omni-model project trained from scratch, with code, data, weights and documentation for listening, seeing and speaking.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Minimind O — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use it when you want a small, inspectable project for learning, training or prototyping multimodal interaction, rather than only consuming a hosted model endpoint.

Who it is for

  • Developers experimenting with compact multimodal models
  • Learners studying end-to-end omni-model training
  • Researchers prototyping audio-and-vision interaction
  • Users wanting a local command-line or WebUI model experiment

Documented uses

  • Explore an omni-model architecture from source
  • Train the documented model using mini or full datasets
  • Run text, audio and visual interaction experiments
  • Test streaming speech, interruption and voice-control workflows described by the project
Before you choose

Requirements & limits

The project is publicly accessible on GitHub, is identified as open source, and uses the Apache-2.0 license. The documented setup requires downloading repository dependencies and model resources.

  • The project combines text, audio and vision inputs with text and streaming-speech output.
  • The main model is described as approximately 0.1B parameters, with a separate larger MoE release also listed.
  • The repository documents both mini and full training data, with the mini path intended for faster experimentation.
  • The implementation uses native PyTorch for core algorithms and supports command-line inference and WebUI demonstrations.
  • The setup depends on several separate encoders and codecs, including SenseVoice-Small, SigLIP2, Mimi and CAMPPlus.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Set up and run Minimind O inference

Prepare the repository resources and launch command-line or WebUI inference.

Before you start

  • Python 3.10 is listed in the documented reference environment.
  • The repository requirements file and the required model resources.
  • A MiniMind language-model weight file in the output directory for the documented setup.

Steps

  1. Clone the repository and install the dependencies from requirements.txt.

  2. Download the SenseVoice-Small, SigLIP2, Mimi, CAMPPlus and MiniMind resources into the documented model and output locations.

  3. Download the released omni weights into the output directory.

  4. Run the documented eval_omni.py command for command-line inference, or copy the Transformers-format model into scripts and start web_demo_omni.py for the WebUI.

Expected outcome

A documented command-line inference path or a non-real-time Gradio WebUI demonstration.

Guide limitations

  • The WebUI demonstration is non-real-time; uploaded or recorded audio is resampled to 16 kHz.
  • The real-time voice-call interface uses a different web_demo.py path.
  • The supplied documentation does not establish runtime performance or deployment interfaces beyond these commands.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
30 September 2026
Original website title
jingyaogong/minimind-o · Official homepage title · retained in its source language
How we review sourcesSuggest a correction