AraT5v2-base-1024-finetune-ar-xlsum

This model is a fine-tuned version of UBC-NLP/AraT5v2-base-1024 on the xlsum dataset. It achieves the following results on the evaluation set:

• Loss: 3.7983
• Rouge-1: 33.4
• Rouge-2: 16.14
• Rouge-l: 29.31
• Gen Len: 18.63
• Bertscore: 74.57

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: 12
• eval_batch_size: 12
• seed: 42
• gradient_accumulation_steps: 16
• total_train_batch_size: 192
• 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
6.1614	1.0	195	3.9898	28.51	12.02	24.64	18.87	72.64
4.5342	2.0	390	3.9048	29.5	13.01	25.85	18.53	73.34
4.2029	3.0	585	3.8162	31.64	14.33	27.54	18.57	73.88
3.9689	4.0	781	3.7949	31.87	14.56	27.9	18.55	74.04
3.8278	5.0	976	3.7702	31.85	14.58	27.74	18.74	73.96
3.6921	6.0	1171	3.7775	32.27	14.95	28.16	18.78	74.23
3.5632	7.0	1367	3.7751	32.54	15.04	28.4	18.72	74.36
3.493	8.0	1562	3.7815	32.35	14.95	28.24	18.71	74.32
3.4189	9.0	1757	3.7908	32.39	14.99	28.32	18.73	74.32
3.3492	9.98	1950	3.7983	32.6	15.19	28.5	18.72	74.35

Framework versions

• Transformers 4.33.1
• Pytorch 2.0.1+cu118
• Datasets 2.14.5
• Tokenizers 0.13.3