helsinki_new_ver1

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

• Loss: 0.4153
• Bleu: 0.0081

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: 20000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu
0.8615	0.5656	1000	0.5171	0.0028
0.809	1.1312	2000	0.4937	0.0035
0.802	1.6968	3000	0.4801	0.0038
0.763	2.2624	4000	0.4691	0.0052
0.7475	2.8281	5000	0.4607	0.0053
0.7363	3.3937	6000	0.4534	0.0057
0.7263	3.9593	7000	0.4470	0.0057
0.7166	4.5249	8000	0.4413	0.0062
0.7119	5.0905	9000	0.4363	0.0061
0.7097	5.6561	10000	0.4324	0.0064
0.6921	6.2217	11000	0.4289	0.0070
0.692	6.7873	12000	0.4258	0.0064
0.6773	7.3529	13000	0.4232	0.0067
0.6918	7.9186	14000	0.4210	0.0073
0.6789	8.4842	15000	0.4194	0.0078
0.6801	9.0498	16000	0.4178	0.0072
0.6734	9.6154	17000	0.4167	0.0070
0.6808	10.1810	18000	0.4159	0.0079
0.669	10.7466	19000	0.4155	0.0081
0.6667	11.3122	20000	0.4153	0.0081

Framework versions

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