m2m100_418M-evaluated-en-to-ar-2000instancesopus-leaningRate2e-05-batchSize16-20epoch-1

This model is a fine-tuned version of facebook/m2m100_418M on the opus100 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3640
• Bleu: 13.1835
• Meteor: 0.1189
• Gen Len: 17.72

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: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Meteor	Gen Len
6.1776	1.0	100	3.8904	10.5866	0.0995	16.64
2.4531	2.0	200	1.0928	12.3452	0.1108	17.0575
0.512	3.0	300	0.3625	10.5224	0.0982	17.2575
0.1924	4.0	400	0.3342	12.4242	0.1098	16.6325
0.1227	5.0	500	0.3403	13.0526	0.1185	17.3475
0.0889	6.0	600	0.3481	13.1323	0.1133	17.815
0.0651	7.0	700	0.3601	12.6684	0.1133	17.3525
0.0533	8.0	800	0.3640	13.1835	0.1189	17.72

Framework versions

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