m2m100_418M-evaluated-en-to-ar-2000instancesUNMULTI-leaningRate2e-05-batchSize8-regu1

This model is a fine-tuned version of facebook/m2m100_418M on the un_multi dataset. It achieves the following results on the evaluation set:

• Loss: 0.3603
• Bleu: 41.8577
• Meteor: 0.4199
• Gen Len: 41.9

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-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 11
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Meteor	Gen Len
5.111	0.5	100	3.2467	29.5017	0.3371	42.425
2.1491	1.0	200	1.0018	33.0563	0.3593	41.205
0.5911	1.5	300	0.4159	34.5818	0.3705	42.0625
0.3546	2.0	400	0.3723	36.6179	0.3823	40.925
0.2487	2.5	500	0.3595	39.0331	0.3956	41.56
0.2365	3.0	600	0.3485	39.5188	0.4023	41.6425
0.1687	3.5	700	0.3542	40.1728	0.4043	42.61
0.1791	4.0	800	0.3466	40.4858	0.4101	41.5575
0.1196	4.5	900	0.3493	41.2457	0.4123	41.755
0.1394	5.0	1000	0.3486	40.5606	0.4114	41.78
0.0958	5.5	1100	0.3568	41.1873	0.4157	41.7275
0.1043	6.0	1200	0.3557	41.2749	0.4165	41.935
0.073	6.5	1300	0.3603	41.8577	0.4199	41.9

Framework versions

• Transformers 4.20.1
• Pytorch 1.11.0
• Datasets 2.1.0
• Tokenizers 0.12.1