The repository includes a dataset and scripts for fine-tuning the GPT-2 model to perform the addition operation task, based on the methodology described in the article "Toolformer: Language Models Can Teach Themselves to Use Tools". The technical report is available here.
To run fine-tuning:
python train.py
The fine-tuned model is available here.
Please download it and place to ./model directory.
To see results:
python evaluate.py
See accuracy scores in results/results_average.json.