AraT5-base-finetune-ar-wikilingua

This model is a fine-tuned version of UBC-NLP/AraT5-base on the wiki_lingua dataset. It achieves the following results on the evaluation set:

• Loss: 4.6110
• Rouge-1: 19.97
• Rouge-2: 6.9
• Rouge-l: 18.25
• Gen Len: 18.45
• Bertscore: 69.44

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0005
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 16
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 250
• num_epochs: 10
• label_smoothing_factor: 0.1

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge-1	Rouge-2	Rouge-l	Gen Len	Bertscore
11.5412	1.0	312	6.8825	5.2	0.69	5.04	19.0	63.2
6.5212	2.0	624	5.8992	8.89	1.4	8.36	17.92	63.9
5.8302	3.0	936	5.3712	9.99	2.21	9.54	15.87	65.08
5.406	4.0	1248	5.0632	13.94	3.5	13.0	15.95	66.83
5.1109	5.0	1560	4.8718	15.28	4.34	14.27	18.26	66.83
4.9004	6.0	1872	4.7631	16.65	4.92	15.46	17.73	67.75
4.754	7.0	2184	4.6920	18.31	5.79	16.9	18.17	68.55
4.6369	8.0	2496	4.6459	18.6	6.12	17.16	18.16	68.66
4.5595	9.0	2808	4.6153	18.94	6.1	17.39	17.82	68.99
4.4967	10.0	3120	4.6110	19.15	6.25	17.55	17.91	69.09

Framework versions

• Transformers 4.19.4
• Pytorch 1.11.0+cu113
• Datasets 2.2.2
• Tokenizers 0.12.1