Arxiv27k T5 Abst Title Gen

🚀 arxiv27k-t5-abst-title-gen/

This model, fine - tuned from mt5 - small on the arxiv - abstract - title dataset, is designed for generating journal titles from abstracts and achieves good Rouge scores on the evaluation set.

🚀 Quick Start

This model is a fine - tuned version of mt5 - small on the arxiv - abstract - title dataset. It achieves the following results on the evaluation set:

• Loss: 1.6002
• Rouge1: 32.8
• Rouge2: 21.9
• Rougel: 34.8

✨ Features

• Can be used for generating journal titles from given abstracts.

📚 Documentation

Model description

Model has been trained with a colab - pro notebook in 4 hours.

Training args

model_args = T5Args()
model_args.max_seq_length = 256
model_args.train_batch_size = 8
model_args.eval_batch_size = 8
model_args.num_train_epochs = 6
model_args.evaluate_during_training = False
model_args.use_multiprocessing = False
model_args.fp16 = False
model_args.save_steps = 40000
model_args.save_eval_checkpoints = False
model_args.save_model_every_epoch = True
model_args.output_dir = OUTPUT_DIR
model_args.no_cache = True
model_args.reprocess_input_data = True
model_args.overwrite_output_dir = True
model_args.num_return_sequences = 1

Framework versions

Contact

• Email: detasar@gmail.com
• Name: Davut Emre Taşar

📄 License

This project is licensed under the Apache 2.0 license.