mizu-gado

This model is a fine-tuned version of roberta-base on the smartbugs dataset available to the public on Kaggle It achieves the following results on the evaluation set:

• Loss: 0.2314
• Accuracy: 0.9397
• F1: 0.9689
• Precision: 0.9397
• Recall: 1.0

Model description

The model has been trained on the solidity smart codes present in the smartbugs wild dataset. USE of robertabase has achieved a high score on the training data.

Intended uses & limitations

The model has put up a huge restriction of 256 max length. This imposition restricts the model to work on most of the contracts. Also, the model has seen a lot of vulnerable training instances and it has been skewed towards the vulnerable class. DOESN'T work on real life data, but has proven itself to be a good auditor for detecting tiny-small vulnerable classes.

Training and evaluation data

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 10
• eval_batch_size: 10
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1	Precision	Recall
0.2357	1.0	3796	0.2351	0.9397	0.9689	0.9397	1.0
0.2292	2.0	7592	0.2325	0.9397	0.9689	0.9397	1.0
0.2396	3.0	11388	0.2314	0.9397	0.9689	0.9397	1.0

Framework versions

• Transformers 4.53.3
• Pytorch 2.6.0+cu124
• Datasets 4.0.0
• Tokenizers 0.21.2