training

This model is a fine-tuned version of huggingface/CodeBERTa-small-v1 on an my a dataset curated from The Technical Debt Dataset.

dataset citation

Valentina Lenarduzzi, Nyyti Saarimäki, Davide Taibi. The Technical Debt Dataset. Proceedings for the 15th Conference on Predictive Models and Data Analytics in Software Engineering. Brazil. 2019.

Model description

Classifies cleaned diffs of code.

• 1: exhibits possible technical debt
• 0: is probably clean

Intended uses & limitations

Limited by many things probably, use with caution. Improvements in progress.

Training and evaluation data

~95% accurate on the test split of dataset above ~.94 F1 score on test split of dataset above.

Training procedure

One epoch of training done on the dataset above.

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 30
• eval_batch_size: 30
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 1

Framework versions

• Transformers 4.35.0
• Pytorch 2.1.0+cu118
• Datasets 2.14.6
• Tokenizers 0.14.1