metadata

base_model: allenai/scibert_scivocab_uncased
tags:
  - generated_from_trainer
  - cybersecurity
metrics:
  - accuracy
model-index:
  - name: my_awesome_model
    results: []
pipeline_tag: text-classification

vuln-cat

This model is a fine-tuned version of allenai/scibert_scivocab_uncased on the None dataset. It achieves the following results on the evaluation set:

Loss: 0.5132
Accuracy: 0.9034

Model description

vuln-cat is a classification model based on fine-tuning of scibert. It categorizes CVE summaries into 11 types of vulnerabilities, with class labels including:

[
    'csrf',
    'directory_traversal',
    'file_inclusion', 
    'input_validation', 
    'memory_corruption', 
    'open_redirect', 
    'overflow', 
    'sql_injection', 
    'ssrf',
    'xss', 
    'xxe'
]

Usage

from transformers import pipeline

text = 'A path traversal exists in a specific dll of Trend Micro Mobile Security (Enterprise) 9.8 SP5 which could allow an authenticated remote attacker to delete arbitrary files.'

classifier = pipeline(
    "text-classification",
    model="conflick0/vuln-cat",
    padding=True,
    truncation=True,
    max_length=512,
)

classifier(text)
# [{'label': 'directory_traversal', 'score': 0.9969494938850403}]

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: 2e-05
train_batch_size: 16
eval_batch_size: 16
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
num_epochs: 8

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
No log	1.0	88	0.3975	0.9006
No log	2.0	176	0.3922	0.9034
No log	3.0	264	0.4732	0.9034
No log	4.0	352	0.5226	0.8949
No log	5.0	440	0.4903	0.9034
0.0513	6.0	528	0.5203	0.9062
0.0513	7.0	616	0.5192	0.8949
0.0513	8.0	704	0.5132	0.9034

Framework versions

Transformers 4.38.2
Pytorch 2.2.1+cu121
Datasets 2.18.0
Tokenizers 0.15.2