helsinki_new_ver2

This model is a fine-tuned version of Helsinki-NLP/opus-mt-en-ZH on the sarahwei/Taiwanese-Minnan-Sutiau dataset. It achieves the following results on the evaluation set:

• Loss: 0.6670
• Bleu: 1.5307

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: 1e-06
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• training_steps: 23000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu
0.8583	0.4230	1000	0.9198	0.1959
0.7889	0.8460	2000	0.8619	0.1895
0.7595	1.2690	3000	0.8263	0.9704
0.7098	1.6920	4000	0.7982	1.0918
0.6963	2.1151	5000	0.7757	1.1072
0.6818	2.5381	6000	0.7568	1.1531
0.6642	2.9611	7000	0.7403	1.2418
0.659	3.3841	8000	0.7262	1.5448
0.6287	3.8071	9000	0.7135	1.3160
0.6251	4.2301	10000	0.7020	1.4177
0.6079	4.6531	11000	0.6918	1.7637
0.6003	5.0761	12000	0.6825	1.3500
0.5874	5.4992	13000	0.6743	1.5090
0.5941	5.9222	14000	0.6670	1.5307

Framework versions

• Transformers 4.51.3
• Pytorch 2.6.0+cu124
• Datasets 3.5.1
• Tokenizers 0.21.1