mBART_skr-en_longerrun

This model was trained from scratch on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.4577
• Bleu: 30.8071
• Gen Len: 34.548

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: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 4
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Gen Len
0.5444	0.72	500	1.3416	28.7505	34.228
0.8576	1.45	1000	1.3411	30.1776	34.328
0.6422	2.18	1500	1.3882	30.2815	34.164
0.532	2.9	2000	1.3716	30.8947	34.556
0.4473	3.63	2500	1.4577	30.8071	34.548

Framework versions

• Transformers 4.21.2
• Pytorch 1.12.1+cu113
• Datasets 2.4.0
• Tokenizers 0.12.1