Update app.py

app.py

@@ -235,7 +235,7 @@ class Translators:
         inputs = tokenizer(prompt, add_special_tokens=False, return_tensors="pt")
         attention_mask = inputs["attention_mask"]        
         input_length = inputs.shape[1]
-        outputs = model.generate(inputs['input_ids'],
+        outputs = model.generate(inputs['input_ids'].to(model.device),
                                 max_new_tokens=512,
                                 early_stopping=True,
                                 num_beams=5,