| Writing logs to /home/ubuntu/buildsCodes/Adversarial_training/trained_models/Multi-delete-our_bert-base-uncased-IMDB/train_log.txt. | |
| Wrote original training args to /home/ubuntu/buildsCodes/Adversarial_training/trained_models/Multi-delete-our_bert-base-uncased-IMDB/training_args.json. | |
| ***** Running training ***** | |
| Num examples = 25000 | |
| Num epochs = 5 | |
| Num clean epochs = 1 | |
| Instantaneous batch size per device = 8 | |
| Total train batch size (w. parallel, distributed & accumulation) = 32 | |
| Gradient accumulation steps = 4 | |
| Total optimization steps = 4410 | |
| ========================================================== | |
| Epoch 1 | |
| Running clean epoch 1/1 | |
| Writing logs to /home/ubuntu/buildsCodes/Adversarial_training/trained_models/Multi-delete-our_bert-base-uncased-IMDB/train_log.txt. | |
| Wrote original training args to /home/ubuntu/buildsCodes/Adversarial_training/trained_models/Multi-delete-our_bert-base-uncased-IMDB/training_args.json. | |
| ***** Running training ***** | |
| Num examples = 25000 | |
| Num epochs = 5 | |
| Num clean epochs = 1 | |
| Instantaneous batch size per device = 8 | |
| Total train batch size (w. parallel, distributed & accumulation) = 32 | |
| Gradient accumulation steps = 4 | |
| Total optimization steps = 4410 | |
| ========================================================== | |
| Epoch 1 | |
| Running clean epoch 1/1 | |
| Writing logs to /home/ubuntu/buildsCodes/Adversarial_training/trained_models/Multi-delete-our_bert-base-uncased-IMDB/train_log.txt. | |
| Wrote original training args to /home/ubuntu/buildsCodes/Adversarial_training/trained_models/Multi-delete-our_bert-base-uncased-IMDB/training_args.json. | |
| ***** Running training ***** | |
| Num examples = 25000 | |
| Num epochs = 5 | |
| Num clean epochs = 1 | |
| Instantaneous batch size per device = 8 | |
| Total train batch size (w. parallel, distributed & accumulation) = 32 | |
| Gradient accumulation steps = 4 | |
| Total optimization steps = 4410 | |
| ========================================================== | |
| Epoch 1 | |
| Running clean epoch 1/1 | |
| Writing logs to /home/ubuntu/buildsCodes/Adversarial_training/trained_models/Multi-delete-our_bert-base-uncased-IMDB/train_log.txt. | |
| Wrote original training args to /home/ubuntu/buildsCodes/Adversarial_training/trained_models/Multi-delete-our_bert-base-uncased-IMDB/training_args.json. | |
| ***** Running training ***** | |
| Num examples = 25000 | |
| Num epochs = 5 | |
| Num clean epochs = 1 | |
| Instantaneous batch size per device = 8 | |
| Total train batch size (w. parallel, distributed & accumulation) = 32 | |
| Gradient accumulation steps = 4 | |
| Total optimization steps = 4410 | |
| ========================================================== | |
| Epoch 1 | |
| Running clean epoch 1/1 | |
| Train accuracy: 97.48% | |
| Eval accuracy: 90.31% | |
| Best score found. Saved model to /home/ubuntu/buildsCodes/Adversarial_training/trained_models/Multi-delete-our_bert-base-uncased-IMDB//best_model/ | |
| ========================================================== | |
| Epoch 2 | |
| Attacking model to generate new adversarial training set... | |
| Total number of attack results: 4403 | |
| Attack success rate: 91.43% [4000 / 4375] | |
| Train accuracy: 98.84% | |
| Eval accuracy: 93.46% | |
| Best score found. Saved model to /home/ubuntu/buildsCodes/Adversarial_training/trained_models/Multi-delete-our_bert-base-uncased-IMDB//best_model/ | |
| ========================================================== | |
| Epoch 3 | |
| Attacking model to generate new adversarial training set... | |
| Writing logs to /home/ubuntu/buildsCodes/Adversarial_training/trained_models/Multi-delete-our_bert-base-uncased-IMDB/train_log.txt. | |
| Wrote original training args to /home/ubuntu/buildsCodes/Adversarial_training/trained_models/Multi-delete-our_bert-base-uncased-IMDB/training_args.json. | |
| ***** Running training ***** | |
| Num examples = 25000 | |
| Num epochs = 5 | |
| Num clean epochs = 1 | |
| Instantaneous batch size per device = 8 | |
| Total train batch size (w. parallel, distributed & accumulation) = 32 | |
| Gradient accumulation steps = 4 | |
| Total optimization steps = 4410 | |
| ========================================================== | |
| Epoch 1 | |
| Running clean epoch 1/1 | |
| Train accuracy: 97.48% | |
| Eval accuracy: 90.31% | |
| Best score found. Saved model to /home/ubuntu/buildsCodes/Adversarial_training/trained_models/Multi-delete-our_bert-base-uncased-IMDB//best_model/ | |
| ========================================================== | |
| Epoch 2 | |
| Attacking model to generate new adversarial training set... | |
| Train accuracy: 98.89% | |
| Eval accuracy: 93.25% | |
| Best score found. Saved model to /home/ubuntu/buildsCodes/Adversarial_training/trained_models/Multi-delete-our_bert-base-uncased-IMDB//best_model/ | |
| ========================================================== | |
| Epoch 3 | |
| Attacking model to generate new adversarial training set... | |
| Total number of attack results: 6088 | |
| Attack success rate: 65.77% [4000 / 6082] | |
| Train accuracy: 70.22% | |
| Eval accuracy: 93.25% | |
| ========================================================== | |
| Epoch 4 | |
| Attacking model to generate new adversarial training set... | |