fd71dba79ce8d49cc852c7361cfedad4

This model is a fine-tuned version of google/mt5-large on the Helsinki-NLP/opus_books [it-ru] dataset. It achieves the following results on the evaluation set:

• Loss: 1.1350
• Data Size: 1.0
• Epoch Runtime: 183.3013
• Bleu: 13.0349

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	23.7266	0	14.1231	0.0116
No log	1	447	21.6652	0.0078	15.7483	0.0091
0.415	2	894	17.0076	0.0156	18.5876	0.0122
0.495	3	1341	13.7980	0.0312	22.7138	0.0114
0.6943	4	1788	9.5620	0.0625	29.2277	0.0105
0.9935	5	2235	6.1368	0.125	40.6361	0.0263
8.0742	6	2682	5.7273	0.25	59.8237	0.0128
2.2073	7	3129	1.4777	0.5	101.9594	7.5941
1.6631	8.0	3576	1.2702	1.0	184.4059	10.0590
1.5121	9.0	4023	1.2029	1.0	187.7614	11.7413
1.3879	10.0	4470	1.1642	1.0	184.0279	12.0998
1.2829	11.0	4917	1.1418	1.0	183.6231	12.4019
1.2278	12.0	5364	1.1284	1.0	184.9419	12.5439
1.1299	13.0	5811	1.1187	1.0	184.3686	12.6358
1.0835	14.0	6258	1.1106	1.0	182.9839	12.8606
1.0406	15.0	6705	1.1055	1.0	183.7175	13.0104
0.9761	16.0	7152	1.1034	1.0	181.0947	12.8474
0.938	17.0	7599	1.1062	1.0	181.6994	12.9996
0.9063	18.0	8046	1.1032	1.0	182.1885	12.8769
0.8546	19.0	8493	1.1186	1.0	184.0141	13.0274
0.8137	20.0	8940	1.1148	1.0	182.5322	13.1157
0.7969	21.0	9387	1.1327	1.0	182.2020	13.0979
0.7312	22.0	9834	1.1350	1.0	183.3013	13.0349

Framework versions

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