opus-mt-lg-en-finetuned-lm-to-en

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

• Loss: 0.3798
• Bleu: 91.6784
• Gen Len: 10.7258

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: 32
• eval_batch_size: 32
• 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	143	2.9523	6.6676	10.7022
No log	2.0	286	2.2556	13.1367	11.2939
No log	3.0	429	1.7848	22.8311	10.5799
2.933	4.0	572	1.4150	32.5314	10.8738
2.933	5.0	715	1.1421	45.7271	10.6075
2.933	6.0	858	0.9408	57.5952	10.6489
1.3478	7.0	1001	0.7880	65.9701	10.716
1.3478	8.0	1144	0.6725	73.9167	10.5937
1.3478	9.0	1287	0.6007	78.7528	10.645
1.3478	10.0	1430	0.5479	81.3326	10.643
0.6179	11.0	1573	0.5037	83.8208	10.6036
0.6179	12.0	1716	0.4726	85.822	10.6489
0.6179	13.0	1859	0.4489	87.3239	10.6529
0.3054	14.0	2002	0.4261	88.3752	10.6844
0.3054	15.0	2145	0.4220	89.6516	10.6923
0.3054	16.0	2288	0.4019	90.5766	10.6805
0.3054	17.0	2431	0.3967	91.0437	10.6982
0.1578	18.0	2574	0.3903	91.0284	10.714
0.1578	19.0	2717	0.3889	90.7559	10.7554
0.1578	20.0	2860	0.3846	91.3402	10.7101
0.0886	21.0	3003	0.3842	91.3279	10.7318
0.0886	22.0	3146	0.3888	91.6286	10.7199
0.0886	23.0	3289	0.3829	91.3329	10.7613
0.0886	24.0	3432	0.3824	91.42	10.7337
0.0536	25.0	3575	0.3849	91.5798	10.7298
0.0536	26.0	3718	0.3809	91.5241	10.7258
0.0536	27.0	3861	0.3810	91.7031	10.7199
0.0392	28.0	4004	0.3807	91.6602	10.712
0.0392	29.0	4147	0.3801	91.6217	10.7199
0.0392	30.0	4290	0.3798	91.6784	10.7258

Framework versions

• Transformers 4.38.2
• Pytorch 2.1.0+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2