68174dedb48e8aee3dbe7e2e374444f6

This model is a fine-tuned version of Helsinki-NLP/opus-mt-en-ru on the Helsinki-NLP/opus_books [fr-no] dataset. It achieves the following results on the evaluation set:

• Loss: 2.7788
• Data Size: 1.0
• Epoch Runtime: 6.5770
• Bleu: 1.8779

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: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• total_train_batch_size: 32
• total_eval_batch_size: 32
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: constant
• num_epochs: 50

Training results

Training Loss	Epoch	Step	Validation Loss	Data Size	Epoch Runtime	Bleu
No log	0	0	8.1576	0	1.1298	0.0120
No log	1	86	7.2886	0.0078	1.6048	0.0364
No log	2	172	6.6430	0.0156	1.4120	0.0139
No log	3	258	6.1895	0.0312	1.6749	0.0290
No log	4	344	5.5434	0.0625	1.9136	0.0593
0.3187	5	430	4.9139	0.125	2.3661	0.1100
1.2067	6	516	4.2736	0.25	3.0119	0.1236
1.5018	7	602	3.7627	0.5	3.9584	0.3964
2.1437	8.0	688	3.3846	1.0	6.2046	0.7190
3.339	9.0	774	3.1913	1.0	5.9831	0.8692
3.1252	10.0	860	3.0527	1.0	6.0412	1.2043
3.0013	11.0	946	2.9586	1.0	5.9485	1.1741
2.798	12.0	1032	2.8913	1.0	6.2445	1.3138
2.6842	13.0	1118	2.8311	1.0	6.0278	1.3558
2.56	14.0	1204	2.7889	1.0	6.3031	1.5464
2.4526	15.0	1290	2.7657	1.0	6.1649	1.5456
2.3505	16.0	1376	2.7463	1.0	6.1921	1.3818
2.2636	17.0	1462	2.7351	1.0	6.1744	1.5696
2.1563	18.0	1548	2.7299	1.0	6.4085	1.7799
2.0449	19.0	1634	2.7390	1.0	6.0078	1.7635
1.9794	20.0	1720	2.7285	1.0	6.2261	1.7834
1.9045	21.0	1806	2.7564	1.0	6.2097	1.6870
1.8111	22.0	1892	2.7386	1.0	6.1472	1.8704
1.7096	23.0	1978	2.7728	1.0	6.1480	1.8485
1.6496	24.0	2064	2.7788	1.0	6.5770	1.8779

Framework versions

• Transformers 4.57.0
• Pytorch 2.8.0+cu128
• Datasets 4.2.0
• Tokenizers 0.22.1