app.py

import gradio as gr
import requests
import os  

def check_hallucination(assertion, citation):
    api_url = "https://api-inference.huggingface.co/models/vectara/hallucination_evaluation_model"
    header = {"Authorization": f"Bearer {hf_token}"}
    payload = {"inputs": f"{assertion} [SEP] {citation}"}

    response = requests.post(api_url, headers=header, json=payload, timeout=120)
    output = response.json()
    output = output[0][0]["score"]

    return f"**hallucination score:** {output}"


# Define the API parameters
vapi_url = "https://api-inference.huggingface.co/models/vectara/hallucination_evaluation_model"

headers = {"Authorization": f"Bearer {hf_token}"}


# Function to query the API
def query(payload):
    response = requests.post(vapi_url, headers=headers, json=payload)
    return response.json()


# Function to evaluate hallucination
def evaluate_hallucination(input1, input2):
    # Combine the inputs
    combined_input = f"{input1}. {input2}"
    
    # Make the API call
    output = query({"inputs": combined_input})
    
    # Extract the score from the output
    score = output[0][0]['score']
    
    # Generate a label based on the score
    if score < 0.5:
        label = f"🔴 High risk. Score: {score:.2f}"
    else:
        label = f"🟢 Low risk. Score: {score:.2f}"
    
    return label

def query_vectara(text):
    user_message = text

    # Read authentication parameters from the .env file
    customer_id = os.getenv('CUSTOMER_ID')
    corpus_id = os.getenv('CORPUS_ID')
    api_key = os.getenv('API_KEY')

    # Define the headers
    api_key_header = {
        "customer-id": customer_id,
        "x-api-key": api_key
    }

    # Define the request body in the structure provided in the example
    request_body = {
        "query": [
            {
                "query": user_message,
                "queryContext": "",
                "start": 1,
                "numResults": 25,
                "contextConfig": {
                    "charsBefore": 0,
                    "charsAfter": 0,
                    "sentencesBefore": 2,
                    "sentencesAfter": 2,
                    "startTag": "%START_SNIPPET%",
                    "endTag": "%END_SNIPPET%",
                },
                "rerankingConfig": {
                    "rerankerId": 272725718,
                    "mmrConfig": {
                        "diversityBias": 0.35
                    }
                },
                "corpusKey": [
                    {
                        "customerId": customer_id,
                        "corpusId": corpus_id,
                        "semantics": 0,
                        "metadataFilter": "",
                        "lexicalInterpolationConfig": {
                            "lambda": 0
                        },
                        "dim": []
                    }
                ],
                "summary": [
                    {
                        "maxSummarizedResults": 5,
                        "responseLang": "auto",
                        "summarizerPromptName": "vectara-summary-ext-v1.2.0"
                    }
                ]
            }
        ]
    }

    # Make the API request using Gradio
    response = requests.post(
        "https://api.vectara.io/v1/query",
        json=request_body,  # Use json to automatically serialize the request body
        verify=True,
        headers=api_key_header
    )

    if response.status_code == 200:
        query_data = response.json()
        if query_data:
            sources_info = []

            # Extract the summary.
            summary = query_data['responseSet'][0]['summary'][0]['text']

            # Iterate over all response sets
            for response_set in query_data.get('responseSet', []):
                # Extract sources
                # Limit to top 5 sources.
                for source in response_set.get('response', [])[:5]:
                    source_metadata = source.get('metadata', [])
                    source_info = {}

                    for metadata in source_metadata:
                        metadata_name = metadata.get('name', '')
                        metadata_value = metadata.get('value', '')

                        if metadata_name == 'title':
                            source_info['title'] = metadata_value
                        elif metadata_name == 'author':
                            source_info['author'] = metadata_value
                        elif metadata_name == 'pageNumber':
                            source_info['page number'] = metadata_value

                    if source_info:
                        sources_info.append(source_info)

            result = {"summary": summary, "sources": sources_info}
            return f"{json.dumps(result, indent=2)}"
        else:
            return "No data found in the response."
    else:
        return f"Error: {response.status_code}"

# Create the Gradio interface
iface = gr.Interface(
    fn=evaluate_hallucination,
    inputs=[gr.Textbox(label="Assertion"), gr.Textbox(label="Citation")],
    outputs=[gr.Label(), gr.Textbox(label="Explanation")],
    live=False,
    title="👋🏻Welcome to 🌟Tonic's 🧠🌈Hallucination Tester 🔴🟢",
    description="How To Use 🌈Hallucination tester: 🗣️📝add any assertion from an LLM or a human 🗣️😷 add any citation from a RAG retriever or a source 👇🏻📩 Press send 🔴red means a 🌈hallucination, 🟢 green means a 🧠credible assertion. Check out the model [vectara/hallucination_evaluation_model](https://huggingface.co/vectara/hallucination_evaluation_model) You can also use 🥒🍆🫑Vectara - Hallucination Tester 🗣️😷 via API below or way by cloning this space. 🧬🔬🔍 Simply click here: Join us : 🌟TeamTonic🌟 is always making cool demos! Join our active builder's🛠️community 👻  [![Join us on Discord](https://img.shields.io/discord/1109943800132010065?label=Discord&logo=discord&style=flat-square)](https://discord.gg/GWpVpekp) On 🤗Huggingface: [TeamTonic](https://huggingface.co/TeamTonic) & [MultiTransformer](https://huggingface.co/MultiTransformer) On 🌐Github: [Tonic-AI](https://github.com/tonic-ai) & contribute to 🌟 [DataTonic](https://github.com/Tonic-AI/DataTonic) 🤗Big thanks to Yuvi Sharma and all the folks at huggingface for the community grant 🤗",
    theme='ParityError/Anime',
)

# Launch the interface
iface.launch()