codet5-small-Generate_Docstrings_for_Python-Condensed

This model is a fine-tuned version of Salesforce/codet5-small on the None dataset. It achieves the following results on the evaluation set:

• Loss: 2.1444
• Rouge1: 0.3828
• Rouge2: 0.2214
• Rougel: 0.3583
• Rougelsum: 0.3661
• Gen Len: 12.6656

Model description

This model is trained to predict the docstring (the output) for a function (the input).

For more information on how it was created, check out the following link: https://github.com/DunnBC22/NLP_Projects/blob/main/Generate%20Docstrings/Smol%20Dataset/Code_T5_Project-Small%20Checkpoint.ipynb

For this model, I trimmed some of the longer samples to quicken the pace of training on consumer hardware.

Intended uses & limitations

This model is intended to demonstrate my ability to solve a complex problem using technology.

Training and evaluation data

Dataset Source: calum/the-stack-smol-python-docstrings (from HuggingFace Datasets; https://huggingface.co/datasets/calum/the-stack-smol-python-docstrings)

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 4

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum	Gen Len
2.9064	1.0	965	2.3096	0.3695	0.2098	0.3464	0.3529	11.7285
2.4836	2.0	1930	2.2051	0.38	0.2176	0.3554	0.3635	12.9401
2.3669	3.0	2895	2.1548	0.3842	0.2219	0.3595	0.3674	13.0029
2.3254	4.0	3860	2.1444	0.3828	0.2214	0.3583	0.3661	12.6656

Framework versions

• Transformers 4.26.1
• Pytorch 1.12.1
• Datasets 2.9.0
• Tokenizers 0.12.1