AquilaX-NL-JSON-Start-Scan
Overview
AquilaX-NL-JSON-Start-Scan is a model built using Hugging Face's T5-small to convert natural language queries about vulnerabilities into JSON queries for MongoDB.
Model Information
Model
- Name: AquilaX-NL-JSON-Start-Scan
- Architecture: T5-small
- Framework: Hugging Face Transformers
Description
The AquilaX-NL-JSON-Start-Scan model is designed to interpret natural language queries related to vulnerabilities in code and convert them into JSON queries that can be executed on a MongoDB database. This facilitates automated scanning and analysis of code repositories for security issues. The model leverages the capabilities of the T5-small architecture, which is well-suited for natural language understanding and generation tasks.
Getting Started
Usage
Below we share some code snippets on how to get quickly started with running the model. First make sure to pip install -U transformers[torch]
, then copy the snippet from the section.
Requirements
pip install transformers[torch]
Inference Code
import json
import requests
def convert_to_json(answer):
"""
Convert a string representation of a dictionary to a JSON object.
This function takes a string representation of a dictionary, cleans it by removing
specific unwanted tokens and correcting boolean representations, and then converts
it into a JSON object.
Parameters:
answer (str): The input string representing a dictionary.
Returns:
dict: The JSON object converted from the input string.
"""
answer = answer.replace("<pad>", "").replace("</s>", "")
answer = answer.strip("'")
answer = answer.replace("false", "False").replace("true", "True")
answer_dict = eval(answer)
answer_json = json.dumps(answer_dict)
json_data = json.loads(answer_json)
return json_data
def valid_url(url):
"""
Validate the given URL against a list of supported platforms.
This function checks if the provided URL belongs to one of the supported
platforms for scanning. If the URL is valid, it returns True. Otherwise,
it returns a message indicating that the URL is not supported and lists the
available scanners.
Parameters:
url (str): The URL to be validated.
Returns:
bool or dict: Returns True if the URL is valid, otherwise returns a
dictionary with a message indicating the URL is not supported
and lists the available scanners.
"""
valid_list = [
"github.com", "bitbucket.org", "sourceforge.net", "aws.amazon.com",
"dev.azure.com", "gitea.com", "gogs.io", "phabricator.com",
"gitkraken.com", "beanstalkapp.com", "gitlab.com"
]
platform = url.split("//")[1].split("/")[0]
if platform in valid_list:
return True
return {
'message': 'Provide a valid URL for scanning. Currently, we support PII_Scanner, SAST_Scanner, Sac_Scanner (Open_Source_Security), IaC_Scanner, Container_Scanner'
}
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import torch
import time
tokenizer = AutoTokenizer.from_pretrained("AquilaX-AI/NL-JSON-Start-Scan")
model = AutoModelForSeq2SeqLM.from_pretrained("AquilaX-AI/NL-JSON-Start-Scan")
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
# Change YOUR_QUERY eg: can this https://github.com/mr-vicky-01/educational-assitant on every week using pii and sast scan
query = "Translate the following text to JSON: " + "YOUR_QUERY".lower()
query = query.replace(",", "")
start = time.time()
inputs = tokenizer(query, return_tensors="pt")
model.to(device)
inputs = inputs.to(device)
outputs = model.generate(**inputs, max_length=256)
answer = tokenizer.decode(outputs[0])
try:
json_data = convert_to_json(answer)
except:
json_data = {'message': 'We encountered an issue with your query. Please use the Personalized Scan option for accurate results.'}
to_return = json_data.copy()
try:
valid = valid_url(json_data["repo"])
if valid != True:
to_return = valid
else:
url = re.findall(r'https?://\S+', query)
to_return['repo'] = url
except:
pass
end = time.time()
print(to_return)
print(f"Time taken: {end - start}")
License
This project is licensed under the Apache 2.0 License - see the LICENSE file for details.
Authors
Acknowledgments
- Hugging Face for the Transformers library.
- Downloads last month
- 645