metadata
tags:
- generated_from_trainer
datasets:
- cnn_dailymail
model-index:
- name: albert_roberta_new_summarization_cnn_dailymail
results: []
albert_roberta_new_summarization_cnn_dailymail
This model is a fine-tuned version of on the cnn_dailymail dataset.
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: 5e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 2000
- num_epochs: 3.0
- mixed_precision_training: Native AMP
Training results
Framework versions
- Transformers 4.12.0.dev0
- Pytorch 1.10.0+cu111
- Datasets 1.18.3
- Tokenizers 0.10.3