Results:

LoRA Model Testing Script

This repository contains a script to test a fine-tuned LoRA model for multimodal (vision-language) tasks.

Setup

1. Install the required dependencies:

   pip install -r requirements.txt
   

2. Make sure you have the LoRA model files in the lora_model directory:

   • adapter_config.json
   • adapter_model.safetensors
   • special_tokens_map.json
   • tokenizer.json
   • tokenizer_config.json

Using the Script

1. Update the image_path variable in test.py to point to your own image file:

   image_path = "your_image.jpg"  # Change this to your image path

2. Run the script:

   python test.py
   

3. Customize the instruction in the script as needed for different types of image analysis:

   instruction = "You are an expert radiographer. Describe accurately what you see in this image."

Hardware Requirements

• For optimal performance, a CUDA-compatible GPU is recommended
• If no GPU is available, the script will fall back to CPU mode (much slower)

Troubleshooting

• If you encounter CUDA out-of-memory errors, you may need to adjust the model loading parameters
• If you have issues with the model loading, ensure the base model is accessible