opus-mt-en-zh-finetuned-audio-product

This model is a fine-tuned version of Helsinki-NLP/opus-mt-en-zh on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.0662
• Bleu: 60.5401

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: 5e-05
• train_batch_size: 64
• eval_batch_size: 64
• 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: cosine
• lr_scheduler_warmup_steps: 100
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu
3.7442	1.0	66	0.1319	37.4394
0.1694	2.0	132	0.0904	48.8190
0.1179	3.0	198	0.0773	51.3552
0.0662	4.0	264	0.0716	51.7823
0.0542	5.0	330	0.0681	54.9762
0.0445	6.0	396	0.0668	55.1506
0.0308	7.0	462	0.0655	56.1614
0.0274	8.0	528	0.0656	57.7665
0.0246	9.0	594	0.0656	58.8837
0.019	10.0	660	0.0654	59.1559
0.017	11.0	726	0.0661	60.0839
0.0151	12.0	792	0.0658	58.4574
0.013	13.0	858	0.0660	59.3195
0.0119	14.0	924	0.0657	59.5030
0.0118	15.0	990	0.0662	60.0147
0.0104	16.0	1056	0.0661	60.6406
0.0101	17.0	1122	0.0662	60.3492
0.01	18.0	1188	0.0662	60.7125
0.0097	19.0	1254	0.0662	60.4941
0.0096	20.0	1320	0.0662	60.5401

Framework versions

• Transformers 4.48.2
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0