97f1f0b808b9d9584c9116f8aa41fe0c

This model is a fine-tuned version of facebook/mbart-large-50-one-to-many-mmt on the Helsinki-NLP/opus_books [es-ru] dataset. It achieves the following results on the evaluation set:

• Loss: 2.6310
• Data Size: 1.0
• Epoch Runtime: 107.1111
• Bleu: 7.0196

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	7.2941	0	9.2743	0.2751
No log	1	419	3.3962	0.0078	10.4124	1.7702
No log	2	838	2.9986	0.0156	11.6560	2.6325
0.0914	3	1257	2.8023	0.0312	14.1871	3.8049
0.0914	4	1676	2.6493	0.0625	17.1266	4.0924
0.1674	5	2095	2.5030	0.125	23.2728	4.9398
0.324	6	2514	2.3695	0.25	35.4740	5.5117
2.1099	7	2933	2.2401	0.5	60.3880	8.8951
1.8339	8.0	3352	2.1387	1.0	108.5625	7.0360
1.5008	9.0	3771	2.1909	1.0	107.3321	7.7451
1.2131	10.0	4190	2.2902	1.0	107.2575	7.9180
0.8869	11.0	4609	2.4522	1.0	106.7648	8.0277
0.7013	12.0	5028	2.6310	1.0	107.1111	7.0196

Framework versions

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