helsinki_new_ver5.2

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: 0.2017
• Bleu: 2.0589
• Ter: 93.3806

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: 3e-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
0.2448	0.4230	1000	0.2484	0.2369	95.6659
0.22	0.8460	2000	0.2238	0.4532	93.8534
0.1998	1.2690	3000	0.2154	0.5837	92.9078
0.1903	1.6920	4000	0.2099	0.8498	92.0410
0.175	2.1151	5000	0.2070	1.1658	93.2230
0.1675	2.5381	6000	0.2051	1.6545	93.2230
0.1684	2.9611	7000	0.2039	1.3913	92.1986
0.1573	3.3841	8000	0.2032	1.5844	92.0410
0.1564	3.8071	9000	0.2027	1.0087	104.3341
0.1532	4.2301	10000	0.2018	2.1594	92.9866
0.1486	4.6531	11000	0.2019	2.2898	93.8534
0.1576	5.0761	12000	0.2018	2.0100	92.4350
0.1439	5.4992	13000	0.2016	2.0117	93.3806
0.1486	5.9222	14000	0.2017	2.0589	93.3806

Framework versions

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