SEBIS/code_trans_t5_small_commit_generation_multitask_finetune

The CodeTrans model is based on the t5-small model architecture . It has its own SentencePiece vocabulary model . It is then fine-tuned on the git commit message generation task for the java commit changes . It can be used on unparsed and untokenized commit changes. It works best with tokenized git commit . The model was trained using the encoder-decoder architecture. The optimizer used is AdaFactor with inverse square root learning rate schedule for pre-training. It has a total of approximately 220M parameters and was trained on a single TPU Pod V3-8 for 500,000 steps in total, using sequence length 512 (batch size,