import streamlit as st import requests import json import os import pandas as pd from sentence_transformers import CrossEncoder import numpy as np import re # Credentials ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ corpus_id = os.environ['VECTARA_CORPUS_ID'] customer_id = os.environ['VECTARA_CUSTOMER_ID'] api_key = os.environ['VECTARA_API_KEY'] # Get Data +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ def get_post_headers() -> dict: """Returns headers that should be attached to each post request.""" return { "x-api-key": api_key, "customer-id": customer_id, "Content-Type": "application/json", } def query_vectara(query: str, filter_str="", lambda_val=0.0) -> str: corpus_key = { "customerId": customer_id, "corpusId": corpus_id, "lexicalInterpolationConfig": {"lambda": lambda_val}, } if filter_str: corpus_key["metadataFilter"] = filter_str data = { "query": [ { "query": query, "start": 0, "numResults": 10, "contextConfig": { "sentencesBefore": 2, "sentencesAfter": 2 }, "corpusKey": [corpus_key], "summary": [ { "responseLang": "eng", "maxSummarizedResults": 5, "summarizerPromptName": "vectara-summary-ext-v1.2.0" }, ] } ] } response = requests.post( "https://api.vectara.io/v1/query", headers=get_post_headers(), data=json.dumps(data), timeout=130, ) if response.status_code != 200: st.error(f"Query failed (code {response.status_code}, reason {response.reason}, details {response.text})") return "" result = response.json() answer = result["responseSet"][0]["summary"][0]["text"] return re.sub(r'\[\d+(,\d+){0,5}\]', '', answer) # 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 # Create the main app with three tabs tab1, tab2, tab3 = st.tabs(["Synthetic Data", "Data Query", "HHEM-Victara Query Tuner"]) with tab1: st.header("Synthetic Data") st.link_button("Create Synthetic Medical Data", "https://chat.openai.com/g/g-XyHciw52w-synthetic-clinical-data") with tab2: st.header("Data Query") st.link_button("Query & Summarize Data", "https://chat.openai.com/g/g-9tWqg4gRY-explore-summarize-medical-data") with tab3: st.header("HHEM-Victara Query Tuner") # User inputs query = st.text_area("Enter your text for query tuning", "", height=75) 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.")