Billwzl's picture
update model card README.md
7025069
metadata
license: apache-2.0
tags:
  - generated_from_trainer
model-index:
  - name: 20split_dataset_version1
    results: []

20split_dataset_version1

This model is a fine-tuned version of distilbert-base-uncased on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 2.1942

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: 64
  • eval_batch_size: 64
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • num_epochs: 12

Training results

Training Loss Epoch Step Validation Loss
2.7475 1.0 11851 2.5194
2.5528 2.0 23702 2.4191
2.4649 3.0 35553 2.3646
2.4038 4.0 47404 2.3289
2.3632 5.0 59255 2.2922
2.3273 6.0 71106 2.2739
2.2964 7.0 82957 2.2494
2.2732 8.0 94808 2.2217
2.2526 9.0 106659 2.2149
2.2369 10.0 118510 2.2029
2.222 11.0 130361 2.2020
2.2135 12.0 142212 2.1942

Framework versions

  • Transformers 4.20.1
  • Pytorch 1.12.0+cu113
  • Datasets 2.3.2
  • Tokenizers 0.12.1