# UbiquitousLearning/mllm

Fast Multimodal LLM on Mobile Devices

Repository: https://github.com/UbiquitousLearning/mllm
Canonical: https://ross.abutalabs.com/products/mllm
Homepage: https://ubiquitouslearning.github.io/mllm/
Language: C++
License: MIT
License Family: permissive
Topics: llama, multimodal, ai, llm, mobile
Last push: 2026-08-19T13:16:34+00:00

## Health v2 (maintenance only)
Score: 73/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 98, release rhythm 39, longevity 78
- inputs: {"age_days": 1099, "days_push": 14, "days_rel": 198, "gap_med": null, "n_releases_24m": 1}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1593, forks 211 (observed 2026-08-28T04:05:08.890766+00:00)

## What it is
MLLM is a fast, lightweight multimodal LLM inference engine written in C++ for mobile and edge devices, with backends for ARM CPU, Qualcomm QNN NPU, OpenCL, CUDA (Jetson), and Ascend NPU. It includes a Python runtime (pymllm) supporting quantized serving of models like Qwen3 and Qwen3-VL.

## Use cases
- run multimodal LLMs on Android phones
- on-device LLM inference on mobile devices
- run Qwen3-VL on Jetson Orin with quantization
- deploy LLM inference on NPU with QNN
- serve quantized LLMs on edge devices
- run vision-language models offline on mobile

## When to choose
- you need fast multimodal LLM inference on mobile or edge hardware
- you target Qualcomm NPU, Ascend, or Jetson devices
- you want C++ on-device inference with quantization support

## When to avoid
- you need a mature general-purpose server LLM stack for datacenter GPUs
- you only need standard CPU inference on desktops where llama.cpp suffices
- you need broad model coverage beyond supported families like Qwen

## Facets
- artifact type: library
- maturity: active
- function: llm-inference, machine-learning, deep-learning, sdk, cli
- domain: large-language-models, artificial-intelligence, mobile-development, embedded-systems, gpu-computing
- platform: cross-platform, cpp, python, embedded
- tags: multimodal-llm, on-device-inference, edge-ai, npu, qnn, quantization, jetson, ascend, qwen, arm-cpu, android, ios, gpu, mobile

## Member repositories
- UbiquitousLearning/mllm (main) score 73

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:08.890766+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 2026-08-30T03:54:14.617996+00:00, confidence not recorded.
  - readme: https://github.com/UbiquitousLearning/mllm (fetched 2026-08-28T04:05:08.890766+00:00, sha f8fb46726e49)
  - homepage: https://ubiquitouslearning.github.io/mllm/ (fetched 2026-08-29T11:25:02.310669+00:00, sha 7dbd8f474a8d)
- Data as of 2026-08-30T08:39:29.467469+00:00.