Update rag_engine.py

rag_engine.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 import html
 import re
+import os
 import threading
 from dataclasses import dataclass
 import faiss
@@ -208,15 +209,13 @@ _LOCAL_MODEL_LOCK = threading.Lock()
 
 def get_embedding_device() -> str:
     """
-    Use the NVIDIA GPU when available
-
-    Fall back to CPU (automatically, if GPU not ready)
+     embedding device
     """
 
-    if torch.cuda.is_available():
-        return "cuda"
+    if os.getenv("SPACES_ZERO_GPU"):
+        return "cpu"
 
-    return "cpu"
+    return "cuda" if torch.cuda.is_available() else "cpu"
 
 
 def get_embedder() -> SentenceTransformer: