654b455562868ebb43c8ffbbdcf93af8

This model is a fine-tuned version of Helsinki-NLP/opus-mt-en-sv on the Helsinki-NLP/opus_books [fi-pl] dataset. It achieves the following results on the evaluation set:

• Loss: 2.9797
• Data Size: 1.0
• Epoch Runtime: 5.2421
• Bleu: 0.5555

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	7.7814	0	1.0052	0.0389
No log	1	70	7.1461	0.0078	1.1038	0.0443
No log	2	140	6.5727	0.0156	1.2015	0.0520
No log	3	210	6.0948	0.0312	1.4516	0.0859
No log	4	280	5.6510	0.0625	1.7623	0.1150
No log	5	350	5.1473	0.125	2.1433	0.1097
No log	6	420	4.5406	0.25	2.7053	0.1446
0.7638	7	490	4.0379	0.5	3.8352	0.2215
3.9138	8.0	560	3.6326	1.0	5.8445	0.2848
3.6252	9.0	630	3.3872	1.0	5.2817	0.2180
3.3105	10.0	700	3.2420	1.0	5.3844	0.3219
3.152	11.0	770	3.1513	1.0	5.4999	0.4649
3.0548	12.0	840	3.0785	1.0	5.5812	0.4815
2.8563	13.0	910	3.0162	1.0	5.6663	0.5489
2.7598	14.0	980	2.9825	1.0	5.6292	0.5155
2.6416	15.0	1050	2.9533	1.0	5.6456	0.4671
2.5251	16.0	1120	2.9414	1.0	5.0727	0.5435
2.471	17.0	1190	2.9317	1.0	5.4107	0.5019
2.3469	18.0	1260	2.9275	1.0	5.2289	0.5205
2.2792	19.0	1330	2.9259	1.0	5.4788	0.5319
2.1728	20.0	1400	2.9446	1.0	5.6755	0.5505
2.0618	21.0	1470	2.9635	1.0	5.3168	0.5170
2.0375	22.0	1540	2.9726	1.0	5.2259	0.5588
1.9011	23.0	1610	2.9797	1.0	5.2421	0.5555

Framework versions

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