Update main.py

main.py

@@ -1,77 +1,38 @@
+# 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
-import threading
 
-# --- Cache Configuration ---
 os.environ["HF_HOME"] = "/app/hf_cache"
 os.environ["TRANSFORMERS_CACHE"] = "/app/hf_cache"
 os.environ["TORCH_DISABLE_CUDA"] = "1"
 
-# --- Paths ---
 MODEL_PATH = './muril_combined_multilingual_model'
 CSV_PATH = './muril_multilingual_dataset.csv'
 HF_REPO = "Sp2503/muril-dataset"
 HF_FILE = "answer_embeddings.pt"
 
-# --- FastAPI Setup ---
-app = FastAPI(title="MuRIL Multilingual QA API")
+print("⚙️ Loading model and embeddings...")
 
-# Global variables
-model = None
-df = None
-answer_embeddings = None
-is_model_ready = False
-loading_lock = threading.Lock()
+# Load model
+model = SentenceTransformer(MODEL_PATH)
+df = pd.read_csv(CSV_PATH).dropna(subset=['question', 'answer'])
 
-# --- Helper: Load embeddings from Hugging Face ---
-def load_embeddings():
-    print("📥 Downloading embeddings from Hugging Face...")
-    hf_path = hf_hub_download(
-        repo_id=HF_REPO,
-        filename=HF_FILE,
-        repo_type="dataset",
-        cache_dir="/tmp"
-    )
-    print(f"✅ Embeddings available at {hf_path}")
-    return torch.load(hf_path, map_location="cpu")
+# Load embeddings from HF
+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")
 
-# --- Resource Loader ---
-def load_resources():
-    global model, df, answer_embeddings, is_model_ready
-    with loading_lock:
-        if is_model_ready:
-            return
-        try:
-            print("⚙️ Loading model and dataset...")
-            model = SentenceTransformer(MODEL_PATH)
-            df = pd.read_csv(CSV_PATH).dropna(subset=['question', 'answer'])
-            answer_embeddings = load_embeddings()
-            is_model_ready = True
-            print("✅ Model and embeddings ready.")
-        except Exception as e:
-            print(f"❌ Error loading resources: {e}")
+print("✅ Model and embeddings loaded.")
 
-# --- Health Check ---
-@app.get("/healthz")
-def health_check():
-    # Always return 200 for Cloud Run health checks
-    return {"status": "ok", "model_loaded": is_model_ready}
+from fastapi import FastAPI
+from pydantic import BaseModel
+from langdetect import detect
 
-# --- Root Endpoint ---
-@app.get("/")
-def root():
-    return {
-        "status": "✅ MuRIL QA API running",
-        "model_loaded": is_model_ready
-    }
+app = FastAPI(title="MuRIL QA API")
 
-# --- Request Models ---
 class QueryRequest(BaseModel):
     question: str
     lang: str = None
@@ -79,15 +40,12 @@ class QueryRequest(BaseModel):
 class QAResponse(BaseModel):
     answer: str
 
-# --- Question Answer Endpoint ---
+@app.get("/")
+def root():
+    return {"status": "✅ Running", "model_loaded": True}
+
 @app.post("/get-answer", response_model=QAResponse)
 def get_answer_endpoint(request: QueryRequest):
-    if not is_model_ready:
-        # Lazy-load the model if first request
-        load_resources()
-        if not is_model_ready:
-            return {"answer": "⏳ Model still loading, please try again shortly."}
-
     question_text = request.question.strip()
     lang_filter = request.lang or detect(question_text)
 
@@ -106,8 +64,3 @@ 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 directly ---
-if __name__ == "__main__":
-    import uvicorn
-    uvicorn.run("main:app", host="0.0.0.0", port=8080)