337719f0b1055befb4fa6e8faaf161b0

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

• Loss: 1.8724
• Data Size: 1.0
• Epoch Runtime: 254.8527
• Bleu: 16.6663

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	7.1696	0	21.1332	0.3852
No log	1	1000	3.3659	0.0078	24.2227	2.6761
No log	2	2000	2.8663	0.0156	25.7403	4.0320
No log	3	3000	2.5638	0.0312	29.7493	5.2350
0.0973	4	4000	2.3327	0.0625	37.2264	6.3076
2.2386	5	5000	2.1332	0.125	52.1154	7.4675
0.1224	6	6000	1.9326	0.25	80.4804	8.2163
0.1519	7	7000	1.7626	0.5	138.4985	9.9221
1.4724	8.0	8000	1.6162	1.0	253.8726	13.4497
1.2135	9.0	9000	1.6199	1.0	256.1921	13.7045
0.9789	10.0	10000	1.6404	1.0	251.4336	12.6859
0.7939	11.0	11000	1.7527	1.0	252.8237	24.0654
0.6297	12.0	12000	1.8724	1.0	254.8527	16.6663

Framework versions

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