t5-finetuned-ar-to-arsl3

This model is a fine-tuned version of PRAli22/arat5-arabic-dialects-translation on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3891
• Bleu: 0.9525

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: 0.0001
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu
No log	0.99	78	0.4271	0.9497
No log	1.99	157	0.3818	0.9485
No log	2.99	236	0.3428	0.9526
No log	3.99	315	0.3173	0.9527
No log	5.0	394	0.3344	0.9518
No log	6.0	473	0.3455	0.9536
0.0868	7.0	552	0.3598	0.9525
0.0868	8.0	631	0.3672	0.9543
0.0868	8.99	709	0.3574	0.9537
0.0868	9.99	788	0.3664	0.9505
0.0868	10.99	867	0.3750	0.9529
0.0868	11.99	946	0.3805	0.9538
0.054	13.0	1025	0.3885	0.9541
0.054	14.0	1104	0.3874	0.9525
0.054	14.83	1170	0.3891	0.9525

Framework versions

• Transformers 4.39.3
• Pytorch 2.1.2
• Datasets 2.18.0
• Tokenizers 0.15.2