mt5-base-arabic

This model is a fine-tuned version of google/mt5-base on arabic subset on the xlsum dataset. It achieves the following results on the evaluation set:

• Loss: 3.2742
• Rouge-1: 22.86
• Rouge-2: 10.31
• Rouge-l: 20.85
• Gen Len: 19.0
• Bertscore: 71.52

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: 8
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 5
• label_smoothing_factor: 0.1

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge-1	Rouge-2	Rouge-l	Gen Len	Bertscore
4.2331	1.0	1172	3.5051	18.54	6.63	16.77	19.0	70.28
3.7075	2.0	2344	3.3737	19.99	7.94	18.19	19.0	70.79
3.5132	3.0	3516	3.3171	20.76	8.57	18.96	19.0	70.95
3.3859	4.0	4688	3.2811	21.49	8.99	19.51	19.0	71.19
3.3012	5.0	5860	3.2742	21.79	9.18	19.77	19.0	71.25

Framework versions

• Transformers 4.18.0
• Pytorch 1.11.0+cu113
• Datasets 2.1.0
• Tokenizers 0.12.1