helsinki_new_ver5.0

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: 0.1331
• Bleu: 24.8564
• Ter: 51.5767

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: 8e-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: 15000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Ter
0.3275	0.5656	1000	0.2091	1.0420	80.6027
0.2661	1.1312	2000	0.1750	8.8029	63.5599
0.2414	1.6968	3000	0.1592	14.6322	59.3833
0.2165	2.2624	4000	0.1512	17.7486	55.9776
0.2079	2.8281	5000	0.1457	19.9557	54.1135
0.1953	3.3937	6000	0.1433	21.4108	53.4688
0.188	3.9593	7000	0.1410	21.9998	52.9643
0.1804	4.5249	8000	0.1379	22.9829	52.6279
0.1797	5.0905	9000	0.1367	23.1706	52.3196
0.1745	5.6561	10000	0.1355	23.3668	51.8290
0.1662	6.2217	11000	0.1343	24.1466	51.7169
0.1691	6.7873	12000	0.1342	24.4332	51.9972
0.1631	7.3529	13000	0.1335	24.4473	51.6608
0.1646	7.9186	14000	0.1330	24.6473	51.6748
0.1599	8.4842	15000	0.1331	24.8564	51.5767

Framework versions

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