use gpu for inf

app.py

@@ -467,7 +467,7 @@ def load_model():
         model = AutoModelForCausalLM.from_pretrained(
             MODEL_ID,
             torch_dtype=torch.float16,
-            device_map="cpu",
+            device_map="auto",
         )
         model.eval()
     return model, tokenizer
@@ -488,7 +488,7 @@ def get_ai_action(game: TinyHanabiGame) -> str:
 
     # Generate
     text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
-    inputs = tokenizer(text, return_tensors="pt")
+    inputs = tokenizer(text, return_tensors="pt").to(model.device)
 
     with torch.no_grad():
         outputs = model.generate(