app.py

from fastapi import FastAPI, HTTPException, Request
from fastapi.responses import JSONResponse, RedirectResponse
from pydantic import BaseModel
from sentence_transformers import SentenceTransformer, util
from transformers import pipeline
from typing import List
import numpy as np

app = FastAPI()

# Load models
model = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
question_model = "deepset/tinyroberta-squad2"
nlp = pipeline('question-answering', model=question_model, tokenizer=question_model)

summarizer = pipeline("summarization", model="facebook/bart-large-cnn")


class ModifyQueryRequest_v3(BaseModel):
    query_string_list: List[str]


class T5QuestionRequest(BaseModel):
    context: str

class T5Response(BaseModel):
    answer: str

# API endpoints
@app.post("/modify_query")
async def modify_query(request: Request):
    try:
        raw_data = await request.json()
        binary_embeddings = model.encode([raw_data['query_string']], precision="binary")
        return JSONResponse(content={'embeddings':binary_embeddings[0].tolist()})
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

@app.post("/modify_query_v3")
async def modify_query_v3(request: Request):
    try:
        # Generate embeddings for a list of query strings
        raw_data = await request.json()
        embeddings = model.encode(raw_data['query_string_list'])
        return JSONResponse(content={'embeddings':[emb.tolist() for emb in embeddings]})
    except Exception as e:
        raise HTTPException(status_code=500, detail=f"Error in modifying query v3: {str(e)}")

@app.post("/answer_question")
async def answer_question(request: Request):
    try:
        raw_data = await request.json()
        res_locs = []
        context_string = ''
        corpus_embeddings = model.encode(raw_data['context'], convert_to_tensor=True)
        query_embeddings = model.encode(raw_data['question'], convert_to_tensor=True)
        hits = util.semantic_search(query_embeddings, corpus_embeddings)
        
        # Collect relevant contexts
        for hit in hits[0]:
            if hit['score'] > 0.4:
                loc = hit['corpus_id']
                res_locs.append(raw_data['locations'][loc])
                context_string += raw_data['context'][loc] + ' '
        
        # If no relevant contexts are found
        if not res_locs:
            answer = "Sorry, I couldn't find any results for your query. Please try again!"
        else:
            # Use the question-answering pipeline
            QA_input = {
                'question': raw_data['question'],
                'context': context_string.replace('\n', ' ')
            }
            result = nlp(QA_input)
            answer = result['answer']
        
        return JSONResponse(content={'answer':answer, "location":res_locs})
    except Exception as e:
        raise HTTPException(status_code=500, detail=f"Error in answering question: {str(e)}")

@app.post("/t5answer", response_model=T5Response)
async def t5answer(request: T5QuestionRequest):
    try:
        # Summarize the context
        response = summarizer(request.context, max_length=130, min_length=30, do_sample=False)
        return T5Response(answer=response[0]["summary_text"])
    except Exception as e:
        raise HTTPException(status_code=500, detail=f"Error in T5 summarization: {str(e)}")

if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=8000)