88fa6a8707203ed1cdba313637b52638

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

• Loss: 2.6923
• Data Size: 1.0
• Epoch Runtime: 173.7785
• Bleu: 9.4933

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	6.3371	0	15.0227	0.5035
No log	1	688	3.4730	0.0078	16.5389	1.9692
No log	2	1376	3.1337	0.0156	18.8642	2.5928
No log	3	2064	2.9232	0.0312	21.9068	3.0416
0.1115	4	2752	2.7608	0.0625	26.6132	3.7047
0.2155	5	3440	2.6156	0.125	36.2651	4.3756
2.4815	6	4128	2.4724	0.25	55.7100	5.2911
2.252	7	4816	2.3335	0.5	95.7813	5.7762
2.028	8.0	5504	2.2318	1.0	174.5723	6.9755
1.6953	9.0	6192	2.2427	1.0	173.3482	6.8769
1.4103	10.0	6880	2.3379	1.0	173.9185	8.9522
1.1403	11.0	7568	2.4994	1.0	173.7038	8.0726
0.9231	12.0	8256	2.6923	1.0	173.7785	9.4933

Framework versions

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