12-18-finetuned-eng-to-para

This model is a fine-tuned version of Tri1/6-12-finetuned-eng-to-para on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0339
• Bleu: 35.7233
• Gen Len: 22.56

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: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 6

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Gen Len
0.0441	1.0	6250	0.0420	31.1327	22.08
0.0412	2.0	12500	0.0397	32.419	22.72
0.0396	3.0	18750	0.0371	35.4708	20.96
0.0385	4.0	25000	0.0356	34.6037	21.12
0.0377	5.0	31250	0.0343	35.7663	22.24
0.0381	6.0	37500	0.0339	35.7233	22.56

Framework versions

• Transformers 4.30.1
• Pytorch 2.0.0
• Datasets 2.1.0
• Tokenizers 0.13.3