app.py

import streamlit as st
import requests
import json
import os
import pandas as pd
from sentence_transformers import CrossEncoder
import numpy as np

# Initialize the HHEM model
model = CrossEncoder('vectara/hallucination_evaluation_model')

# Function to compute HHEM scores
def compute_hhem_scores(texts, summary):
    pairs = [[text, summary] for text in texts]
    scores = model.predict(pairs)
    return scores

# Define the Vectara query function
def vectara_query(query: str, config: dict):
    corpus_key = [{
        "customerId": config["customer_id"],
        "corpusId": config["corpus_id"],
        "lexicalInterpolationConfig": {"lambda": config.get("lambda_val", 0.5)},
    }]
    data = {
        "query": [{
            "query": query,
            "start": 0,
            "numResults": config.get("top_k", 10),
            "contextConfig": {
                "sentencesBefore": 2,
                "sentencesAfter": 2,
            },
            "corpusKey": corpus_key,
            "summary": [{
                "responseLang": "eng",
                "maxSummarizedResults": 5,
            }]
        }]
    }

    headers = {
        "x-api-key": config["api_key"],
        "customer-id": config["customer_id"],
        "Content-Type": "application/json",
    }
    response = requests.post(
        headers=headers,
        url="https://api.vectara.io/v1/query",
        data=json.dumps(data),
    )
    if response.status_code != 200:
        st.error(f"Query failed (code {response.status_code}, reason {response.reason}, details {response.text})")
        return [], ""

    result = response.json()
    responses = result["responseSet"][0]["response"]
    summary = result["responseSet"][0]["summary"][0]["text"]

    res = [[r['text'], r['score']] for r in responses]
    return res, summary

# Streamlit UI setup
st.title("Vectara Content Query Interface")

# User inputs
query = st.text_input("Enter your query here", "")
lambda_val = st.slider("Lambda Value", min_value=0.0, max_value=1.0, value=0.5)
top_k = st.number_input("Top K Results", min_value=1, max_value=50, value=10)

if st.button("Query Vectara"):
    config = {
        "api_key": os.environ.get("VECTARA_API_KEY", ""),
        "customer_id": os.environ.get("VECTARA_CUSTOMER_ID", ""),
        "corpus_id": os.environ.get("VECTARA_CORPUS_ID", ""),
        "lambda_val": lambda_val,
        "top_k": top_k,
    }

    results, summary = vectara_query(query, config)

    if results:
        st.subheader("Summary")
        st.write(summary)
        
        st.subheader("Top Results")
        
        # Extract texts from results
        texts = [r[0] for r in results[:5]]
        
        # Compute HHEM scores
        scores = compute_hhem_scores(texts, summary)
        
        # Prepare and display the dataframe
        df = pd.DataFrame({'Fact': texts, 'HHEM Score': scores})
        st.dataframe(df)
    else:
        st.write("No results found.")