PontifexMaximus/Arabic2

opus-mt-ar-en-finetuned-ar-to-en

This model is a fine-tuned version of Helsinki-NLP/opus-mt-ar-en on the opus_infopankki dataset. It achieves the following results on the evaluation set:

• Loss: 0.7636
• Bleu: 53.5086
• Gen Len: 13.5728

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: 2e-06
• train_batch_size: 64
• eval_batch_size: 64
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Gen Len
No log	1.0	278	1.5114	35.2767	14.2084
1.6677	2.0	556	1.4025	37.5243	14.0245
1.6677	3.0	834	1.3223	39.4262	13.8101
1.4743	4.0	1112	1.2567	40.7045	13.8533
1.4743	5.0	1390	1.2001	41.8356	13.8083
1.3428	6.0	1668	1.1504	43.2448	13.6958
1.3428	7.0	1946	1.1072	44.177	13.6783
1.2595	8.0	2224	1.0701	45.17	13.6587
1.1829	9.0	2502	1.0345	45.9612	13.6706
1.1829	10.0	2780	1.0042	46.9009	13.6236
1.1188	11.0	3058	0.9760	47.7478	13.6205
1.1188	12.0	3336	0.9505	48.3082	13.6283
1.0735	13.0	3614	0.9270	48.9782	13.6203
1.0735	14.0	3892	0.9060	49.5541	13.6311
1.0269	15.0	4170	0.8869	49.9905	13.6222
1.0269	16.0	4448	0.8700	50.4806	13.6047
0.9983	17.0	4726	0.8538	50.9186	13.6159
0.9647	18.0	5004	0.8398	51.3492	13.6146
0.9647	19.0	5282	0.8271	51.7219	13.5275
0.9398	20.0	5560	0.8156	52.0177	13.5756
0.9398	21.0	5838	0.8053	52.3619	13.5807
0.9206	22.0	6116	0.7963	52.6051	13.5652
0.9206	23.0	6394	0.7885	52.8322	13.5669
0.9012	24.0	6672	0.7818	52.9402	13.5701
0.9012	25.0	6950	0.7762	53.1182	13.5695
0.8965	26.0	7228	0.7717	53.1596	13.5612
0.8836	27.0	7506	0.7681	53.3116	13.5719
0.8836	28.0	7784	0.7656	53.4399	13.5758
0.8777	29.0	8062	0.7642	53.4805	13.5737
0.8777	30.0	8340	0.7636	53.5086	13.5728

Framework versions

• Transformers 4.20.1
• Pytorch 1.12.0
• Datasets 2.3.2
• Tokenizers 0.12.1