app.py

import os
import torch
import pandas as pd
from fastapi import FastAPI
from pydantic import BaseModel
from sentence_transformers import SentenceTransformer, util
from huggingface_hub import snapshot_download

# --- Cache Config ---
os.environ["HF_HOME"] = "/app/hf_cache"
os.environ["TRANSFORMERS_CACHE"] = "/app/hf_cache"
os.environ["TORCH_DISABLE_CUDA"] = "1"

# --- Download Model & Embeddings from Hub ---
HF_REPO = "Sp2503/Muril-Model"

print("📦 Downloading model & embeddings from Hugging Face Hub...")
model_dir = snapshot_download(repo_id=HF_REPO, repo_type="model")
print(f"✅ Model snapshot available at: {model_dir}")

MODEL_PATH = model_dir
CSV_PATH = os.path.join(model_dir, "muril_multilingual_dataset.csv")
EMBED_PATH = os.path.join(model_dir, "answer_embeddings.pt")

# --- Load Model ---
print("⚙️ Loading model and embeddings...")
model = SentenceTransformer(MODEL_PATH)
df = pd.read_csv(CSV_PATH).dropna(subset=['question', 'answer'])
answer_embeddings = torch.load(EMBED_PATH, map_location="cpu")
print("✅ Model and embeddings loaded successfully.")

# --- FastAPI App ---
app = FastAPI(title="MuRIL Multilingual QA API")

class QueryRequest(BaseModel):
    question: str
    lang: str = None

class QAResponse(BaseModel):
    answer: str

@app.get("/")
def root():
    return {"status": "✅ API is running", "model_loaded": True}

@app.post("/get-answer", response_model=QAResponse)
def get_answer_endpoint(request: QueryRequest):
    question_text = request.question.strip()
    lang_filter = request.lang

    filtered_df = df
    filtered_embeddings = answer_embeddings
    if 'lang' in df.columns and lang_filter:
        mask = df['lang'] == lang_filter
        filtered_df = df[mask].reset_index(drop=True)
        filtered_embeddings = answer_embeddings[mask.values]

    if len(filtered_df) == 0:
        return {"answer": f"No data found for language '{lang_filter}'."}

    question_emb = model.encode(question_text, convert_to_tensor=True)
    cosine_scores = util.pytorch_cos_sim(question_emb, filtered_embeddings)
    best_idx = torch.argmax(cosine_scores).item()
    answer = filtered_df.iloc[best_idx]['answer']
    return {"answer": answer}