CodeTrans model for code documentation generation python

Pretrained model on programming language python using the t5 base model architecture. It was first released in this repository. This model is trained on tokenized python code functions: it works best with tokenized python functions.

Model description

This CodeTrans model is based on the t5-base model. It has its own SentencePiece vocabulary model. It used transfer-learning pre-training on 7 unsupervised datasets in the software development domain. It is then fine-tuned on the code documentation generation task for the python function/method.

Intended uses & limitations

The model could be used to generate the description for the python function or be fine-tuned on other python code tasks. It can be used on unparsed and untokenized python code. However, if the python code is tokenized, the performance should be better.

How to use

Here is how to use this model to generate python function documentation using Transformers SummarizationPipeline:

from transformers import AutoTokenizer, AutoModelWithLMHead, SummarizationPipeline

pipeline = SummarizationPipeline(
    model=AutoModelWithLMHead.from_pretrained("SEBIS/code_trans_t5_base_code_documentation_generation_python_transfer_learning_finetune"),
    tokenizer=AutoTokenizer.from_pretrained("SEBIS/code_trans_t5_base_code_documentation_generation_python_transfer_learning_finetune", skip_special_tokens=True),
    device=0
)

tokenized_code = "def e ( message , exit_code = None ) : print_log ( message , YELLOW , BOLD ) if exit_code is not None : sys . exit ( exit_code )"
pipeline([tokenized_code])

Run this example in colab notebook.

Training data

The supervised training tasks datasets can be downloaded on Link

Training procedure

Transfer-learning Pretraining

The model was trained on a single TPU Pod V3-8 for half million steps in total, using sequence length 512 (batch size 4096). It has a total of approximately 220M parameters and was trained using the encoder-decoder architecture. The optimizer used is AdaFactor with inverse square root learning rate schedule for pre-training.

Fine-tuning

This model was then fine-tuned on a single TPU Pod V2-8 for 2000 steps in total, using sequence length 512 (batch size 256), using only the dataset only containing python code.

Evaluation results

For the code documentation tasks, different models achieves the following results on different programming languages (in BLEU score):

Test results :

Created by Ahmed Elnaggar | LinkedIn and Wei Ding | LinkedIn