This is an AI tool for grading assignments based on given guidance. It uses the Longformer model, a transformer-based model that can handle long texts, from the Hugging Face Transformers library.
Firstly, you need to have Python 3.7 or later installed on your system.
- Clone this repository.
- Install the necessary Python packages. You can do this by running
pip install -r requirements.txt. - Place your assignment guidance in a file called 'guidance.txt' and your assignments in a folder called 'assignments'. The assignments should be in '.txt' format and their corresponding grades should be in a file with the same name, appended with '_grade.txt'. For example, for an assignment 'assignment1.txt', its grade should be in 'assignment1_grade.txt'.
- Run
python train.pyto train the model. This will create a file called 'model.pt', which is the saved trained model. - To test the model on a new assignment, place the assignment in a file called 'test_assignment.txt' and run
python test.py. This will print out the grade the model predicts for the assignment.
Note: The model is currently configured to handle assignments of up to 8,000 words in length. If your assignments are longer than this, you may need to modify the split_into_chunks method in train.py and test.py.
If you want to change the model, tokenizer, or any other aspects of the training/testing process, you can do so in train.py and test.py.
Enjoy grading assignments with AI!
This model is not guaranteed to provide accurate grades and should be used as a tool to assist in grading rather than a replacement for human grading. This is only a test model and should not be deployed for industry use. For more accurate model visit: