Edit model card

cantonese-chinese-translation-pycantonese-v22

This model is a fine-tuned version of fnlp/bart-base-chinese on an unknown dataset. It achieves the following results on the evaluation set:

  • Loss: 0.1513
  • Bleu: 7.2337
  • Chrf: 6.716
  • Gen Len: 8.3247

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: 2e-05
  • train_batch_size: 16
  • eval_batch_size: 16
  • seed: 42
  • gradient_accumulation_steps: 4
  • total_train_batch_size: 64
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • num_epochs: 30
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Bleu Chrf Gen Len
0.475 2.91 5000 0.1728 4.088 3.9621 7.6729
0.2232 5.82 10000 0.1513 7.2337 6.716 8.3247
0.1892 8.73 15000 0.1397 4.7516 4.6853 8.1438
0.1667 11.64 20000 0.1370 3.3546 3.4247 7.9183
0.1506 14.55 25000 0.1327 3.8002 3.7862 8.4654

Framework versions

  • Transformers 4.28.1
  • Pytorch 2.2.2+cu121
  • Datasets 2.18.0
  • Tokenizers 0.13.3
Downloads last month
0
Safetensors
Model size
338M params
Tensor type
F32
·