from sentence_transformers import SentenceTransformer, CrossEncoder, util
import torch
import pickle
import pandas as pd
import gradio as gr
bi_encoder = SentenceTransformer("multi-qa-MiniLM-L6-cos-v1")
cross_encoder = CrossEncoder("cross-encoder/ms-marco-MiniLM-L-6-v2")
corpus_embeddings=pd.read_pickle("corpus_embeddings_cpu.pkl")
corpus=pd.read_pickle("corpus.pkl")
def search(query,top_k=100):
    print("Top 5 Answer by the NSE:")
    print()
    ans=[]
    ##### Sematic Search #####
    # Encode the query using the bi-encoder and find potentially relevant passages
    question_embedding = bi_encoder.encode(query, convert_to_tensor=True)
    hits = util.semantic_search(question_embedding, corpus_embeddings, top_k=top_k)
    hits = hits[0]  # Get the hits for the first query
    ##### Re-Ranking #####
    # Now, score all retrieved passages with the cross_encoder
    cross_inp = [[query, corpus[hit['corpus_id']]] for hit in hits]
    cross_scores = cross_encoder.predict(cross_inp)
    # Sort results by the cross-encoder scores
    for idx in range(len(cross_scores)):
        hits[idx]['cross-score'] = cross_scores[idx]
    hits = sorted(hits, key=lambda x: x['cross-score'], reverse=True)
    
    for idx, hit in enumerate(hits[0:5]):
        ans.append(corpus[hit['corpus_id']])
    return ans[0],ans[1],ans[2],ans[3],ans[4]
iface = gr.Interface(fn=search, inputs=["text"], outputs=["textbox","textbox","textbox","textbox","textbox"],examples=["How big is London?", "Where is Rome?","Who is steve jobs?","What is the most interesting thing about our universe?"],article="This is a semantic search engine powered by SentenceTransformers (Nils_Reimers) with a retrieval and reranking system on Wikipedia corpus. It will show the top 5 results",title="Neural Search Engine").launch()