license	base_model	tags	model-index
apache-2.0	Salesforce/codet5-base	generated_from_keras_callback	name results ITRLTrained

ITRLTrained

This model is a fine-tuned version of Salesforce/codet5-base on ITRLDataset

Model description

The model is a fine-tuned version of the CodeT5-base model by Salesforce. It has been fine-tuned on a custom dataset with over 1200 entries.

Intended uses & limitations

Its intended use is to act as a text-to-code generator. The Natural Language (NL) input is in the form of algorithmic statements and the model generates 1-2 lines of Python code corresponding to the input.

Training and evaluation data

AshArya/ITRLDataset

Training procedure

Fine-tuning of Salesforce/codet5-base using PyTorch and Transformers with custom training arguments and a custom training loop.

Training hyperparameters

The following hyperparameters were used during training:

• optimizer: XLA
• training_precision: float32

Training results

The model achieves high accuracy during human evaluation. It achieves a ROUGE-L score of 0.359662899235267 and a chrF score of 9.637061227631952 after the fine-tuning. The original model achieves a score of 0.06735368389780155 and 3.021597905833814 respectively.

Framework versions

• Transformers 4.35.2
• TensorFlow 2.14.0
• Tokenizers 0.15.0