helsinki_new_ver6.0

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: 0.6805
• Bleu: 0.7794
• Ter: 99.4557

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: 16
• 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: 14000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Ter
1.0468	0.4230	1000	0.9170	0.0299	130.9748
0.905	0.8460	2000	0.8370	0.0696	112.0732
0.8412	1.2690	3000	0.7959	0.2321	106.4819
0.8053	1.6920	4000	0.7646	0.7019	100.2474
0.777	2.1151	5000	0.7416	0.8663	100.4453
0.757	2.5381	6000	0.7222	0.8328	100.2474
0.7492	2.9611	7000	0.7088	0.8269	100.0990
0.716	3.3841	8000	0.6968	0.7207	100.0495
0.7169	3.8071	9000	0.6877	0.8254	99.4062
0.6951	4.2301	10000	0.6805	0.7794	99.4557

Framework versions

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