Update app.py

app.py

@@ -124,8 +124,16 @@ class PersistentSearchEngine:
         self.collection = self.client.get_or_create_collection(name=collection_name)
         
         # 2. Load Models
-        self.bi_encoder = SentenceTransformer('all-MiniLM-L6-v2')
-        self.cross_encoder = CrossEncoder('cross-encoder/ms-marco-MiniLM-L-6-v2')
+        # We force the device to CPU to avoid "meta tensor" errors in Docker
+        device = "cpu" 
+        self.bi_encoder = SentenceTransformer('all-MiniLM-L6-v2', device=device)
+        
+        # We disable "low_cpu_mem_usage" to prevent the model from loading as a ghost (meta device)
+        self.cross_encoder = CrossEncoder(
+            'cross-encoder/ms-marco-MiniLM-L-6-v2', 
+            device=device,
+            automodel_args={"low_cpu_mem_usage": False} 
+        )
         
         # 3. Initialize/Load BM25 (Sparse)
         self.bm25 = None