Pegasus-x-base-govreport-12288-1024-numepoch-10
This model is a fine-tuned version of google/pegasus-x-base on the govreport-summarization dataset. It achieves the following results on the evaluation set:
- Loss: 1.6234
Model description
More information needed
Evaluation Score
'ROUGE':
{
'rouge1': 0.5012,
'rouge2': 0.2205,
'rougeL': 0.2552,
'rougeLsum': 0.2554
}
'BERT_SCORE'
{'f1': 0.859,
'precision': 0.8619,
'recall': 0.8563
}
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: 0.0001
- train_batch_size: 1
- eval_batch_size: 2
- seed: 42
- gradient_accumulation_steps: 64
- total_train_batch_size: 64
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 10
Training results
Training Loss | Epoch | Step | Validation Loss |
---|---|---|---|
2.1149 | 0.37 | 100 | 1.9237 |
1.9545 | 0.73 | 200 | 1.8380 |
1.8835 | 1.1 | 300 | 1.7574 |
1.862 | 1.46 | 400 | 1.7305 |
1.8536 | 1.83 | 500 | 1.7100 |
1.8062 | 2.19 | 600 | 1.6944 |
1.8161 | 2.56 | 700 | 1.6882 |
1.7611 | 2.92 | 800 | 1.6803 |
1.7878 | 3.29 | 900 | 1.6671 |
1.7299 | 3.65 | 1000 | 1.6599 |
1.7636 | 4.02 | 1100 | 1.6558 |
1.7262 | 4.38 | 1200 | 1.6547 |
1.715 | 4.75 | 1300 | 1.6437 |
1.7178 | 5.12 | 1400 | 1.6445 |
1.7163 | 5.48 | 1500 | 1.6386 |
1.7367 | 5.85 | 1600 | 1.6364 |
1.7114 | 6.21 | 1700 | 1.6365 |
1.6452 | 6.58 | 1800 | 1.6309 |
1.7251 | 6.94 | 1900 | 1.6301 |
1.6726 | 7.31 | 2000 | 1.6305 |
1.7104 | 7.67 | 2100 | 1.6285 |
1.6739 | 8.04 | 2200 | 1.6252 |
1.7082 | 8.4 | 2300 | 1.6246 |
1.6888 | 8.77 | 2400 | 1.6244 |
1.6609 | 9.13 | 2500 | 1.6256 |
1.6707 | 9.5 | 2600 | 1.6241 |
1.669 | 9.86 | 2700 | 1.6234 |
Framework versions
- Transformers 4.30.2
- Pytorch 2.0.1+cu117
- Datasets 2.13.1
- Tokenizers 0.13.3
- Downloads last month
- 10
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.