airKlizz's picture
update model card README.md
fb4cd10
metadata
license: apache-2.0
tags:
  - summarization
  - generated_from_trainer
metrics:
  - rouge
model-index:
  - name: mt5-base-wikinewssum-german
    results: []

mt5-base-wikinewssum-german

This model is a fine-tuned version of google/mt5-base on an unknown dataset. It achieves the following results on the evaluation set:

  • Loss: 2.5135
  • Rouge1: 8.0553
  • Rouge2: 2.7846
  • Rougel: 6.2182
  • Rougelsum: 7.6203

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: 5.6e-05
  • train_batch_size: 4
  • eval_batch_size: 4
  • seed: 42
  • gradient_accumulation_steps: 2
  • total_train_batch_size: 8
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • num_epochs: 8

Training results

Training Loss Epoch Step Validation Loss Rouge1 Rouge2 Rougel Rougelsum
No log 1.0 723 2.7112 7.3681 2.3679 5.5705 6.7588
No log 2.0 1446 2.6178 7.8539 2.7551 6.2081 7.4139
No log 3.0 2169 2.5756 7.8401 2.6075 6.0135 7.4303
No log 4.0 2892 2.5465 8.1097 2.8525 6.268 7.6482
3.4589 5.0 3615 2.5315 8.0192 2.7848 6.2484 7.5859
3.4589 6.0 4338 2.5222 8.1063 2.8986 6.337 7.6564
3.4589 7.0 5061 2.5136 8.0565 2.8707 6.2732 7.6105
3.4589 8.0 5784 2.5135 8.0553 2.7846 6.2182 7.6203

Framework versions

  • Transformers 4.13.0
  • Pytorch 1.10.1
  • Datasets 1.16.1
  • Tokenizers 0.10.3