dhchoi/manchu-llama32-11b-vision-mlx

Llama 3.2 11B Vision MLX

This is a MLX-optimized version of the Llama 3.2 11B Vision model, converted for use with Apple Silicon devices.

Model Description

• Model Type: Vision-Language Model
• Architecture: MllamaForConditionalGeneration
• Parameters: 11B
• Framework: MLX (Apple Silicon optimized)
• Base Model: Llama 3.2 11B Vision

Features

• Optimized for Apple Silicon (M1/M2/M3/M4) devices
• Vision and text understanding capabilities
• Efficient inference with MLX framework
• Support for image-text conversations

Usage

import mlx.core as mx
from mlx_vlm import load, generate

# Load the model
model, processor = load("path/to/model")

# Generate text from image and prompt
image_path = "your_image.jpg"
prompt = "Describe this image"

response = generate(model, processor, image_path, prompt)
print(response)

Requirements

• macOS with Apple Silicon
• MLX framework
• MLX-VLM package

Installation

pip install mlx mlx-vlm

Model Details

• Vision Config:

  • Image size: 560x560
  • Patch size: 14
  • Hidden size: 1280
  • Layers: 32

• Text Config:

  • Vocab size: 128,256
  • Hidden size: 4096
  • Layers: 40
  • Attention heads: 32

License

Please refer to the original Llama 3.2 license terms.

Citation

If you use this model, please cite the original Llama 3.2 paper and acknowledge the MLX conversion.