helsinki_new_ver4

This model is a fine-tuned version of Helsinki-NLP/opus-mt-en-ZH on the mozilla-foundation/common_voice_12_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5400
• Bleu: 2.4304

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: 23000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu
0.7027	0.6418	1000	0.6716	2.8141
0.6767	1.2837	2000	0.6546	9.1063
0.6526	1.9255	3000	0.6394	1.9859
0.643	2.5674	4000	0.6252	12.4882
0.6445	3.2092	5000	0.6118	8.8121
0.6326	3.8511	6000	0.6010	12.7405
0.604	4.4929	7000	0.5926	1.4845
0.5877	5.1348	8000	0.5827	12.9972
0.5721	5.7766	9000	0.5753	1.5982
0.5826	6.4185	10000	0.5672	1.6842
0.5622	7.0603	11000	0.5619	14.0609
0.5486	7.7022	12000	0.5557	14.2992
0.5451	8.3440	13000	0.5507	15.4044
0.5571	8.9859	14000	0.5463	8.4964
0.5448	9.6277	15000	0.5422	8.8203
0.5306	10.2696	16000	0.5400	2.4304

Framework versions

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