app.py

# # install correct gradio version
# import os
# os.system("pip uninstall -y gradio")
# os.system("pip install gradio==3.35.2")

import gradio as gr
import urllib.request
from datetime import date
from bs4 import BeautifulSoup

from avidtools.datamodels.report import Report
from avidtools.datamodels.components import *
from avidtools.datamodels.enums import *

# def generate_report():
def generate_report(uri,title,abstract,classof,type,risk_domain,sep,lifecycle):
# def generate_report(scraped_input, selections):
    report = Report()

    # report.affects = Affects(
    #     developer = [],
    #     deployer = ['Hugging Face'],
    #     artifacts = [Artifact(
    #         type = ArtifactTypeEnum.model,
    #         name = model_id
    #     )]
    # )    
    report.problemtype = Problemtype(
        # classof = clas,
        classof = classof,
        type = type,
        description = LangValue(
            lang = 'eng',
            value = title
        )
    )
    report.references = [
        Reference(
            label = title,
            url = uri
        )
    ]
    report.description = LangValue(
        lang = 'eng',
        value = abstract
    )
    report.impact = Impact(
        avid = AvidTaxonomy(
            risk_domain = risk_domain,
            sep_view = sep,
            lifecycle_view = lifecycle,
            taxonomy_version = '0.2'
        )
    )
    report.reported_date = date.today()
    
    return report.dict()

# parses out title and abstract: works only on arxiv or acl
def parse_uri(uri):
    if not uri.startswith("https://arxiv.org") and not uri.startswith("https://aclanthology.org"):
        raise gr.Error("Please supply an arxiv.org or aclanthology.org link!")

    try:
        response = urllib.request.urlopen(uri)
        html = response.read()
        doc = BeautifulSoup(html, features="lxml")
    except:
        raise gr.Error("Cannot fetch information. Please check the link!")

    if uri.startswith("https://arxiv.org"):
        title = doc.find("meta", property="og:title")["content"]
        abstract = doc.find("meta", property="og:description")["content"].replace("\n", "")
    else:
        title = doc.find("meta", property="og:title")["content"]
        abstract = doc.find("div", {"class": "card-body"}).text[8:].replace("\n", "")

    return f"### {title}\n", abstract

def empty_all_fields():
    return None, None, None, None, None, {}

demo = gr.Blocks(theme=gr.themes.Soft())

with demo:

    gr.Markdown("# Report AI Vulnerability Research")
    gr.Markdown("""
    AI Researchers are focusing increasing on evaluating AI models and finding novel failure modes in them. \
    With the deluge of AI research coming up, it's really difficult to track and stay updated on recent papers on AI failures.\
    [AVID](https://avidml.org) makes it easy to do that. This space demonstrates how you can report an interesting research paper \
    (maybe yours!) on LLM vulnerabilities to AVID in a structured manner.
    """)
    with gr.Row():
        with gr.Column(scale=3):
            gr.Markdown("""
            ## Step 1: Paste a link.\n\
            (arXiv or ACL Anthology)
            """)
            input_uri = gr.Text(label="Link",value="Enter URL")
        with gr.Column(scale=8):
                title = gr.Markdown("### ")
                description = gr.Markdown("")

    with gr.Row():
        with gr.Column(scale=3):
            gr.Markdown("""## Step 2: Categorize the research.""")

            classof = gr.Radio(label="Class", choices=[ce.value for ce in ClassEnum])
            type = gr.Radio(label="Type", choices=[te.value for te in TypeEnum])
            lifecycle = gr.CheckboxGroup(label="Lifecycle Categories", choices=[le.value for le in LifecycleEnum])
            risk_domain = gr.CheckboxGroup(label="Risk Domain", choices=['Security','Ethics','Performance'])
            sep = gr.CheckboxGroup(label="Effect Categories", choices=[se.value for se in SepEnum])

        with gr.Column(scale=3):
            gr.Markdown("""
            ## Step 3: Generate a report to submit to AVID.

            The title and abstract get auto-populated from Step 1. The taxonomy categories populate from your selections in Step 2.
            """)
            report_button = gr.Button("Generate Report")
            report_json = gr.Json(label="AVID Report")

    input_uri.change(
        fn=empty_all_fields,
        inputs=[],
        outputs=[classof,type,risk_domain,sep,lifecycle,report_json]
    )
    input_uri.change(
        fn=parse_uri,
        inputs=[input_uri],
        outputs=[title,description]
    )
    report_button.click(
        fn=generate_report,
        inputs=[input_uri,title,description,classof,type,risk_domain,sep,lifecycle],
        outputs=[report_json]
    )

demo.launch()