Update app.py

app.py

@@ -24,13 +24,7 @@ if torch.cuda.is_available():
     model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", load_in_4bit=True)
     tokenizer = AutoTokenizer.from_pretrained(model_id)
     tokenizer.use_default_system_prompt = False
-    query_pipeline1 = Pipeline()
-    query_pipeline1.add_component(
-        "text_embedder",
-        SentenceTransformersTextEmbedder(
-            model="intfloat/multilingual-e5-large", 
-            prefix="query:",
-        ))
+    
 
     
 
@@ -92,6 +86,13 @@ def generate(
                          Firmati alla fine di ogni risposta '-AvvoChat'.'''})
     for user, assistant in chat_history:
         conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
+    query_pipeline1 = Pipeline()
+    query_pipeline1.add_component(
+        "text_embedder",
+        SentenceTransformersTextEmbedder(
+            model="intfloat/multilingual-e5-large", 
+            prefix="query:",
+        ))
     materiali = richiamo_materiali(message, alpha=1.0, n_items=5)
     documenti = ''
     for idx, d in enumerate(materiali):