a3bde9cfb216f6a0e283d05c726335bf

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

• Loss: 3.3954
• Data Size: 1.0
• Epoch Runtime: 12.1830
• Bleu: 11.0567

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	7.5248	0	1.2095	0.5683
No log	1	27	6.7667	0.0078	1.4942	0.6694
No log	2	54	6.4283	0.0156	2.2159	0.6543
No log	3	81	6.1098	0.0312	3.1578	0.8285
No log	4	108	5.6630	0.0625	5.1736	1.0561
No log	5	135	5.0443	0.125	6.0862	1.9067
No log	6	162	4.3176	0.25	7.8561	2.7855
No log	7	189	3.7650	0.5	9.3579	3.7672
0.8934	8.0	216	3.3040	1.0	12.4846	5.2690
0.8934	9.0	243	3.1375	1.0	11.9947	6.3657
2.5324	10.0	270	3.1435	1.0	13.3137	6.7718
2.5324	11.0	297	3.2093	1.0	14.1662	7.9532
1.5287	12.0	324	3.3088	1.0	15.4696	11.8808
0.9017	13.0	351	3.3954	1.0	12.1830	11.0567

Framework versions

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