opus_mt_en_zh_AIchallenger

This model is a fine-tuned version of opus-mt-en-zh on the AIChallenger2017 dataset. It achieves the following results on the evaluation set:

• Loss: 1.9379
• Bleu: 65.1489
• Meteor: 0.1558
• Gen Len: 12.4552

Model description

More information needed

Intended uses & limitations

This model is used to run the translation model on the client side.

Training and evaluation data

Dataset Source: https://tianchi.aliyun.com/dataset/174937

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 64
• eval_batch_size: 64
• seed: 2024
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 5.0

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Meteor	Gen Len
2.2886	1.0	156250	2.0967	65.1489	0.1576	12.8416
2.1316	2.0	312500	2.0353	65.1489	0.1591	12.7059
2.05	3.0	468750	1.9878	65.1489	0.1612	12.7766
1.9868	4.0	625000	1.9567	65.1489	0.1619	12.7332
1.9345	5.0	781250	1.9379	65.1489	0.1615	12.7575

Framework versions

• Transformers 4.45.2
• Pytorch 2.5.0+cu124
• Datasets 3.0.1
• Tokenizers 0.20.1