f2a254ca87a674f770e3482b904a7336

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

• Loss: 2.9077
• Data Size: 1.0
• Epoch Runtime: 5.7557
• Bleu: 0.9303

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	7.9852	0	1.0313	0.0187
No log	1	75	7.0896	0.0078	1.8169	0.0216
No log	2	150	6.4901	0.0156	1.2675	0.0123
No log	3	225	6.0555	0.0312	1.4153	0.0123
No log	4	300	5.5600	0.0625	1.5753	0.0789
No log	5	375	4.9630	0.125	1.9565	0.0373
No log	6	450	4.4028	0.25	2.7072	0.0346
No log	7	525	3.9807	0.5	3.6940	0.0418
3.7798	8.0	600	3.6331	1.0	6.3021	0.1063
3.5666	9.0	675	3.4253	1.0	5.5565	0.2843
3.3259	10.0	750	3.3041	1.0	5.5804	0.3805
3.1983	11.0	825	3.2043	1.0	5.9144	0.4566
3.0369	12.0	900	3.1334	1.0	5.5680	0.5230
2.9113	13.0	975	3.0688	1.0	6.0732	0.5838
2.814	14.0	1050	3.0244	1.0	5.6112	0.6339
2.6864	15.0	1125	2.9779	1.0	5.9027	0.6517
2.5893	16.0	1200	2.9578	1.0	5.7051	0.6521
2.4946	17.0	1275	2.9415	1.0	5.9149	0.7250
2.4246	18.0	1350	2.9288	1.0	5.6919	0.7447
2.3156	19.0	1425	2.9326	1.0	5.8616	0.7507
2.2499	20.0	1500	2.9000	1.0	5.9752	0.7759
2.154	21.0	1575	2.9075	1.0	6.4975	0.8442
2.0921	22.0	1650	2.9200	1.0	5.8360	0.8208
1.9875	23.0	1725	2.9155	1.0	5.6037	0.9272
1.93	24.0	1800	2.9077	1.0	5.7557	0.9303

Framework versions

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