rjxai-bert model card

Overview

This model is a fine-tuned version of bert-base-chinese on the rjxdataset-textclassification-chinese-46000 dataset.

Dataset

DatasetDict({
    train: Dataset({
        features: ['label', 'text', 'input_ids', 'token_type_ids', 'attention_mask'],
        num_rows: 36800
    })
    valid: Dataset({
        features: ['label', 'text', 'input_ids', 'token_type_ids', 'attention_mask'],
        num_rows: 4600
    })
    test: Dataset({
        features: ['label', 'text', 'input_ids', 'token_type_ids', 'attention_mask'],
        num_rows: 4600
    })
})

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 1

Training results

eval dataset results

{‘accuracy’: 0.9997826086956522}

Training time return

Training seconds: 4017