Update app.py

app.py

@@ -29,7 +29,7 @@ def build_chain():
     vectorstore = FAISS.from_documents(docs, embeddings)
     retriever = vectorstore.as_retriever(search_kwargs={"k": 3})
 
-    model_id = "Qwen/Qwen2.5-0.5B-Instruct"
+    model_id = "Qwen/Qwen2.5-2.5B-Instruct"
     tokenizer = AutoTokenizer.from_pretrained(model_id)
 
     # sigurnosno: ako nema pad token
@@ -38,7 +38,7 @@ def build_chain():
 
     model = AutoModelForCausalLM.from_pretrained(
         model_id,
-        device_map="cpu",          # menjaš u "auto" ako imaš GPU space
+        device_map="auto",          # menjaš u "auto" ako imaš GPU space
         torch_dtype=torch.float32
     )