app.py

from fastapi import FastAPI, HTTPException
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")

# Request models
class ModifyQueryRequest(BaseModel):
    query_string: str

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

class AnswerQuestionRequest(BaseModel):
    question: str
    context: List[str]
    locations: List[str]

class T5QuestionRequest(BaseModel):
    context: str

# Response models
class ModifyQueryResponse(BaseModel):
    embeddings: List[List[float]]

class AnswerQuestionResponse(BaseModel):
    answer: str
    locations: List[str]

class T5Response(BaseModel):
    answer: str

# API endpoints
@app.post("/modify_query", response_model=ModifyQueryResponse)
async def modify_query(request: ModifyQueryRequest):
    try:
        # Generate embeddings
        embeddings = model.encode([request.query_string])
        return ModifyQueryResponse(embeddings=[emb.tolist() for emb in embeddings])
    except Exception as e:
        raise HTTPException(status_code=500, detail=f"Error in modifying query: {str(e)}")

@app.post("/modify_query_v3", response_model=ModifyQueryResponse)
async def modify_query_v3(request: ModifyQueryRequest_v3):
    try:
        # Generate embeddings for a list of query strings
        embeddings = model.encode(request.query_string_list)
        return ModifyQueryResponse(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", response_model=AnswerQuestionResponse)
async def answer_question(request: AnswerQuestionRequest):
    try:
        res_locs = []
        context_string = ''
        corpus_embeddings = model.encode(request.context, convert_to_tensor=True)
        query_embeddings = model.encode(request.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(request.locations[loc])
                context_string += request.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': request.question,
                'context': context_string.replace('\n', ' ')
            }
            result = nlp(QA_input)
            answer = result['answer']
        
        return AnswerQuestionResponse(answer=answer, locations=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)