e1f4adcacf0d04f75b86407e4c7f2acd

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

• Loss: 2.0471
• Data Size: 1.0
• Epoch Runtime: 110.7223
• Bleu: 11.3063

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	6.9406	0	9.6447	1.2111
No log	1	434	5.2292	0.0078	11.7176	3.6834
No log	2	868	4.8065	0.0156	12.7087	5.8623
No log	3	1302	3.3408	0.0312	14.7806	5.9133
No log	4	1736	2.5304	0.0625	17.7199	4.7952
0.15	5	2170	1.9733	0.125	24.1405	7.2875
1.8619	6	2604	1.7953	0.25	37.2570	8.6292
1.6065	7	3038	1.6582	0.5	62.1927	13.0228
1.3509	8.0	3472	1.5848	1.0	112.2284	21.0278
0.9848	9.0	3906	1.6312	1.0	109.9129	19.1991
0.7195	10.0	4340	1.7470	1.0	112.8104	20.8269
0.5481	11.0	4774	1.9236	1.0	111.5746	12.5108
0.4113	12.0	5208	2.0471	1.0	110.7223	11.3063

Framework versions

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