qingy2024

Extract Mllama vision encoder from unsloth/Llama-3.2-11B-Vision

621808b verified 27 days ago

602 Bytes

library_name: transformers
base_model: unsloth/Llama-3.2-11B-Vision
tags:
  - mllama
  - vision
  - vision-encoder

Llama-3.2-11B Vision Encoder

This repository contains only the MllamaVisionModel weights extracted from unsloth/Llama-3.2-11B-Vision. It intentionally excludes the text decoder and language model weights.

Load it with:

from transformers import MllamaVisionModel

vision = MllamaVisionModel.from_pretrained("cminst/Llama-3.2-11B-VisionEncoder")

For LEGATO configs, set encoder_pretrained_model_name_or_path to:

cminst/Llama-3.2-11B-VisionEncoder