This is the source code for our IEEE Access publication MILE: Memory-Interactive Learning Engine for Neuro-Symbolic Solutions to Mathematical Problems.
The code was written and verified on Python 3.11 and PyTorch 2.1.0.
The other required libraries include :
- json
- numpy
- transformers
To start the training, run:
python ./train.py
To switch to other encoder and decoder models, modify the encode_method and decode_method in config.py, or assign @encode_method="..." @decode_method="..." in the training command:
python ./train.py @encode_method="..." @decode_method="..."