4e35f9982b19120d9477e44328095c1c

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

• Loss: 3.3937
• Data Size: 1.0
• Epoch Runtime: 28.0215
• Bleu: 4.2573

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	5.9971	0	2.4639	0.8074
No log	1	88	5.4109	0.0078	3.0426	0.8957
No log	2	176	4.8928	0.0156	4.7072	1.0249
No log	3	264	4.5026	0.0312	6.5184	1.1337
No log	4	352	3.9560	0.0625	8.4220	1.6097
No log	5	440	3.6070	0.125	9.9703	2.2041
0.2965	6	528	3.2516	0.25	12.5379	2.5741
1.052	7	616	2.9820	0.5	16.3498	3.4256
2.4333	8.0	704	2.7443	1.0	28.2953	4.0755
1.9088	9.0	792	2.7747	1.0	28.7679	4.2844
1.3674	10.0	880	2.8805	1.0	25.3317	4.9430
1.0102	11.0	968	3.1215	1.0	26.7665	4.3768
0.7053	12.0	1056	3.3937	1.0	28.0215	4.2573

Framework versions

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