# rhymes-ai/Aria

Codebase for Aria - an Open Multimodal Native MoE

Repository: https://github.com/rhymes-ai/Aria
Canonical: https://ross.abutalabs.com/products/aria
Language: Jupyter Notebook
License: Apache-2.0
License Family: permissive
Topics: mixture-of-experts, multimodal, vision-and-language
Last push: 2025-01-22T03:25:37+00:00

## Health v2 (maintenance only)
Score: 23/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 2, release rhythm 35, longevity 50
- inputs: {"age_days": 704, "days_push": 588, "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 1087, forks 89 (observed 2026-08-28T04:03:32.287788+00:00)

## What it is
Aria is an open multimodal native Mixture-of-Experts (MoE) model with 25.3B total parameters (3.9B activated per token) and a 64K multimodal context window. This repository provides the inference and fine-tuning codebase, with model weights hosted on Hugging Face.

## Use cases
- run inference on a multimodal vision-language model
- understand documents and videos with an LLM
- fine-tune a mixture-of-experts model cheaply
- build a chatbot that handles images and text
- process long multimodal contexts up to 64K tokens
- compare open MoE models for multimodal tasks

## When to choose
- you need strong video and document understanding in an open model
- you want fast inference via sparse MoE activation
- you have an A100-class GPU and want a single-GPU multimodal model
- you want Apache-2.0 licensed weights and fine-tuning code

## When to avoid
- you only have consumer GPUs with limited VRAM
- you need text-only language modeling at minimal cost
- you want a fully managed API rather than self-hosted weights

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, llm-inference, llm-training, nlp, image-processing, video-processing
- domain: large-language-models, deep-learning, machine-learning, computer-vision
- platform: python
- tags: mixture-of-experts, multimodal, vision-language-model, open-weights, fine-tuning, transformers, natural-language-processing, gpu, linux

## Member repositories
- rhymes-ai/Aria (main) score 23

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:32.287788+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 2026-08-30T06:49:29.082242+00:00, confidence not recorded.
  - readme: https://github.com/rhymes-ai/Aria (fetched 2026-08-28T04:03:32.287788+00:00, sha 73c16d0ed265)
- Data as of 2026-08-30T08:39:29.467469+00:00.