# JIA-Lab-research/MGM

Official repo for "Mini-Gemini: Mining the Potential of Multi-modality Vision Language Models"

Repository: https://github.com/JIA-Lab-research/MGM
Canonical: https://ross.abutalabs.com/products/mgm
Language: Python
License: Apache-2.0
License Family: permissive
Topics: generation, large-language-models, vision-language-model
Last push: 2024-05-04T14:36:51+00:00

## Health v2 (maintenance only)
Score: 25/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 0, release rhythm 35, longevity 63
- inputs: {"age_days": 890, "days_push": 851, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 3327, forks 276 (observed 2026-08-28T04:07:56.488457+00:00)

## What it is
Official PyTorch implementation of Mini-Gemini, a multimodal vision-language model framework built on LLaVA that supports dense and MoE LLMs from 2B to 34B. It combines dual vision encoders for high-resolution image understanding, reasoning, and image generation in a single model.

## Use cases
- run a vision language model for image understanding and reasoning
- generate images from text and image prompts
- fine-tune a multimodal LLM on custom data
- evaluate vision-language models on benchmarks
- try a high-resolution image chat demo
- compare dense and MoE multimodal models

## When to choose
- you need a research-grade VLM with both image understanding and generation
- you want to build on or extend LLaVA-style architectures
- you need high-resolution visual reasoning with dual encoders
- you want pretrained checkpoints from 2B to 34B including LLaMA3-based variants

## When to avoid
- you need a production-ready, commercially supported multimodal API
- you only need text-only LLM inference
- you lack a GPU or cannot install flash-attention
- you need lightweight on-device inference

## Facets
- artifact type: library
- maturity: maintenance
- function: llm-inference, llm-training, machine-learning, image-processing, chatbot
- domain: large-language-models, computer-vision, deep-learning, artificial-intelligence
- platform: python
- tags: vision-language-model, multimodal, image-generation, image-understanding, llava, research-code, moe, gpu, linux, docker

## Member repositories
- JIA-Lab-research/MGM (main) score 25

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:07:56.488457+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 2026-08-29T18:41:45.718819+00:00, confidence not recorded.
  - readme: https://github.com/JIA-Lab-research/MGM (fetched 2026-08-28T04:07:56.488457+00:00, sha 3313f6fb5b4f)
- Data as of 2026-08-30T08:39:29.467469+00:00.