ed3c7accf3e9eef3114691b8b7b520c5

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

• Loss: 2.6903
• Data Size: 1.0
• Epoch Runtime: 95.8963
• Bleu: 7.7780

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	6.8156	0	8.3446	0.2649
No log	1	367	3.9618	0.0078	10.2047	2.3114
No log	2	734	3.3378	0.0156	10.5312	3.0801
No log	3	1101	3.0832	0.0312	12.4524	3.7396
No log	4	1468	2.8969	0.0625	16.2080	4.4253
0.1601	5	1835	2.7389	0.125	22.1757	5.1719
2.65	6	2202	2.5719	0.25	34.3092	5.6794
2.3821	7	2569	2.4171	0.5	54.9511	6.3453
2.0768	8.0	2936	2.2921	1.0	98.5353	7.0496
1.7404	9.0	3303	2.2960	1.0	96.7490	7.5943
1.4095	10.0	3670	2.3902	1.0	95.1176	8.8112
1.1454	11.0	4037	2.5435	1.0	95.9685	10.5792
0.9371	12.0	4404	2.6903	1.0	95.8963	7.7780

Framework versions

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