.

create_app.py

@@ -32,7 +32,8 @@ def load_models():
         QWEN_TOKENIZER = AutoTokenizer.from_pretrained(model_name, device='auto')
         QWEN_TOKENIZER.pad_token_id = QWEN_TOKENIZER.eos_token_id
         print("QWEN TOKENIZER LOADED")
-        QWEN_MODEL = AutoModelForCausalLM.from_pretrained(model_name, device='auto').half()
+        QWEN_MODEL = AutoModelForCausalLM.from_pretrained(model_name).half()
+        QWEN_MODEL = QWEN_MODEL.to(device)
         print("QWEN MODEL LOADED")
         MODELS_LOADED = True
     print("LOAD ENDED")