mbart-large-50-finetuned-ba-to-en

This model is a fine-tuned version of facebook/mbart-large-50-many-to-many-mmt on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 4.3077
• Bleu: 3.5599
• Gen Len: 23.974

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: 2e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 0
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• training_steps: 15000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Gen Len
1.8324	0.6949	1000	3.2741	2.0466	33.962
1.4095	1.3899	2000	3.2448	2.8318	28.761
1.2485	2.0848	3000	3.2848	3.5036	24.234
1.0617	2.7797	4000	3.3263	2.9476	27.438
0.9345	3.4746	5000	3.3947	3.5117	26.216
0.836	4.1696	6000	3.5736	3.3995	26.529
0.7317	4.8645	7000	3.5429	3.2189	28.059
0.6277	5.5594	8000	3.8036	3.2916	25.754
0.5748	6.2543	9000	3.9252	3.4899	24.827
0.5136	6.9493	10000	3.8955	3.2303	24.422
0.4363	7.6442	11000	4.0975	3.1755	24.992
0.4042	8.3391	12000	4.1533	3.564	24.6
0.3726	9.0341	13000	4.2092	3.5126	24.042
0.3311	9.7290	14000	4.2611	3.4499	24.355
0.3166	10.4239	15000	4.3077	3.5599	23.974

Framework versions

• Transformers 4.46.3
• Pytorch 2.4.0
• Datasets 3.1.0
• Tokenizers 0.20.3