Update main.py

main.py

@@ -1,15 +1,18 @@
-# main.py
 import os
 import torch
 import pandas as pd
 from fastapi import FastAPI
+from pydantic import BaseModel
 from sentence_transformers import SentenceTransformer, util
+from langdetect import detect
 from huggingface_hub import hf_hub_download
 
+# --- Cache configuration ---
 os.environ["HF_HOME"] = "/app/hf_cache"
 os.environ["TRANSFORMERS_CACHE"] = "/app/hf_cache"
-os.environ["TORCH_DISABLE_CUDA"] = "1"
+os.environ["TORCH_DISABLE_CUDA"] = "1"  # CPU only
 
+# --- Paths ---
 MODEL_PATH = './muril_combined_multilingual_model'
 CSV_PATH = './muril_multilingual_dataset.csv'
 HF_REPO = "Sp2503/muril-dataset"
@@ -17,22 +20,27 @@ HF_FILE = "answer_embeddings.pt"
 
 print("⚙️ Loading model and embeddings...")
 
-# Load model
+# Load MuRIL model
 model = SentenceTransformer(MODEL_PATH)
+
+# Load CSV dataset
 df = pd.read_csv(CSV_PATH).dropna(subset=['question', 'answer'])
 
-# Load embeddings from HF
-hf_path = hf_hub_download(repo_id=HF_REPO, filename=HF_FILE, repo_type="dataset", cache_dir="/tmp")
+# Load precomputed embeddings from Hugging Face
+hf_path = hf_hub_download(
+    repo_id=HF_REPO,
+    filename=HF_FILE,
+    repo_type="dataset",
+    cache_dir="/tmp"
+)
 answer_embeddings = torch.load(hf_path, map_location="cpu")
 
-print("✅ Model and embeddings loaded.")
-
-from fastapi import FastAPI
-from pydantic import BaseModel
-from langdetect import detect
+print("✅ Model and embeddings loaded successfully.")
 
-app = FastAPI(title="MuRIL QA API")
+# --- FastAPI app ---
+app = FastAPI(title="MuRIL Multilingual QA API")
 
+# --- Request/Response models ---
 class QueryRequest(BaseModel):
     question: str
     lang: str = None
@@ -40,10 +48,12 @@ class QueryRequest(BaseModel):
 class QAResponse(BaseModel):
     answer: str
 
+# --- Root endpoint ---
 @app.get("/")
 def root():
-    return {"status": "✅ Running", "model_loaded": True}
+    return {"status": "✅ Running MuRIL QA API", "model_loaded": True}
 
+# --- QA endpoint ---
 @app.post("/get-answer", response_model=QAResponse)
 def get_answer_endpoint(request: QueryRequest):
     question_text = request.question.strip()
@@ -64,3 +74,8 @@ def get_answer_endpoint(request: QueryRequest):
     best_idx = torch.argmax(cosine_scores).item()
     answer = filtered_df.iloc[best_idx]['answer']
     return {"answer": answer}
+
+# --- Run app ---
+if __name__ == "__main__":
+    import uvicorn
+    uvicorn.run("main:app", host="0.0.0.0", port=8080)