Update ai_engine.py

ai_engine.py

@@ -16,7 +16,10 @@ from huggingface_hub import login, hf_hub_download
 
 API_KEY = os.getenv("OPENROUTER_API_KEY")
 MODEL = os.getenv("OPENROUTER_MODEL", "google/gemma-2-9b-it:free")
-
+global _embed_model
+if _embed_model is None:
+    from sentence_transformers import SentenceTransformer
+    _embed_model = SentenceTransformer('all-MiniLM-L6-v2')
 
 class LocalModelHandler:
     def __init__(self, repo_id, device=None, use_quantization=False):
@@ -120,10 +123,7 @@ class LocalModelHandler:
 _embed_model = None
 
 def get_embedding(text):
-    global _embed_model
-    if _embed_model is None:
-        from sentence_transformers import SentenceTransformer
-        _embed_model = SentenceTransformer('all-MiniLM-L6-v2')
+
     return _embed_model.encode(text).tolist()
 
 def chat_stream(messages):