moht24/qwen3-vl-dr-lora

Qwen3-VL DR LoRA Adapter

Model description

This repository contains the LoRA adapter weights for a Qwen3-VL-based diabetic retinopathy classification setup.

This adapter is intended to be loaded on top of the base model:

unsloth/Qwen3-VL-8B-Instruct-unsloth-bnb-4bit

The adapter is part of the Afterimage project and is used for retinal fundus image analysis.

Base model

• unsloth/Qwen3-VL-8B-Instruct-unsloth-bnb-4bit

Intended use

This adapter is intended for research, educational, and prototype inference use for diabetic retinopathy classification from retinal fundus images.

Limitations

• This adapter is not a standalone model and requires the compatible base model.
• This system is not a medical device.
• This system is not currently able to replace clinical judgment.
• Results may vary depending on preprocessing, prompting, image quality, and dataset differences.

Training / fine-tuning

This adapter was fine-tuned as part of the Afterimage project for diabetic retinopathy classification.

Input

• Retinal fundus images
• Prompt format expected by the associated inference pipeline

Output

• Diabetic retinopathy classification response

Files

Typical adapter files include:

• best_lora/adapter_config.json
• best_lora/adapter_model.safetensors

Authors

Afterimage project

Disclaimer

This repository is provided for research and academic use only.