Apollo-VL-Edge-3B (MLX FP16)
Intelligence isn't about scale. It's about precision.
Overview
Apollo-VL-Edge-3B (MLX FP16) is the native Apple Silicon port of
Apollo-VL-Edge-3B, engineered by
Pluto AI Labs. This repository contains the
MLX FP16 quantization, converted from the original FP16 Transformers checkpoint and optimized to run natively on
Apple M-Series chips (M1/M2/M3/M4) using the unified memory architecture via the
mlx-vlm framework.
This is the full-precision MLX variant — recommended for maximum reasoning quality on Apple Silicon when RAM permits.
Specifications
| Property | Value |
|---|
| Base Model | Pluto-AI-Labs/Apollo-VL-Edge-3B |
| Architecture | Qwen2.5-VL |
| Format | MLX FP16 |
| File Size | ~7.5 GB |
| Recommended RAM | 16 GB (M-Series Mac) |
| Target Hardware | Apple M1/M2/M3/M4 (16GB+) |
| Precision Loss | Baseline |
Quick Start
Install the mlx-vlm framework and run inference locally:
```bash
pip install mlx-vlm
```
```bash
python -m mlx_vlm.generate
--model Pluto-AI-Labs/Apollo-VL-Edge-3B-MLX-FP16
--image ./your_image.png
--prompt "Analyze this image step-by-step."
--max-tokens 512
```
Python API
```python
from mlx_vlm import load, generate
from mlx_vlm.prompt_utils import apply_chat_template
from mlx_vlm.utils import load_config
model_path = "Pluto-AI-Labs/Apollo-VL-Edge-3B-MLX-FP16"
model, processor = load(model_path)
config = load_config(model_path)
prompt = apply_chat_template(
processor,
config,
"Analyze this image step-by-step.",
num_images=1
)
output = generate(
model,
processor,
"./your_image.png",
prompt,
max_tokens=512,
verbose=True
)
```
MLX Quantization Family
All Apollo-VL MLX variants are available for different hardware targets:
💡 For the original FP16 Transformers checkpoint, GGUF quantizations, and full benchmark results, see the
main model repository.
About Pluto AI Labs
Pluto AI Labs is an independent open-source AI research lab focused on efficient intelligence, multimodal reasoning, model distillation, and edge deployment. We explore how capable AI systems can be made smaller, faster, and more accessible without requiring hyperscale infrastructure.
Citation
If you use Apollo-VL-Edge-3B in your research or projects, please cite:
```bibtex
@misc{apollo_vl_edge_3b,
title = {Apollo-VL-Edge-3B: Elite Visual Reasoning on Edge Hardware},
author = {Siddharth N.R. and Pluto AI Labs},
year = {2026},
howpublished = {Hugging Face},
url = {
https://huggingface.co/Pluto-AI-Labs/Apollo-VL-Edge-3B}
}
```
Apollo-VL-Edge-3B · MLX FP16
Precision over scale. Reasoning at the edge.
Built by Pluto AI Labs
