outputs
This model is a fine-tuned version of facebook/bart-large on the cnn_dailymail 3.0.0 dataset.
Model description
More information needed
Intended uses & limitations
This is a work in progress. Please don't use these weights.
Training and evaluation data
More information needed
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.0001
- train_batch_size: 1
- eval_batch_size: 2
- seed: 42
- distributed_type: IPU
- gradient_accumulation_steps: 256
- total_train_batch_size: 2048
- total_eval_batch_size: 4
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: constant
- num_epochs: 2.0
- training precision: Mixed Precision
Training results
Framework versions
- Transformers 4.18.0
- Pytorch 1.10.0+cpu
- Datasets 2.2.1
- Tokenizers 0.12.1