cuda fix

app.py

@@ -3,6 +3,8 @@ import gradio as gr
 from transformers import GPT2LMHeadModel, GPT2Tokenizer
 tokenizer = GPT2Tokenizer.from_pretrained("sberbank-ai/mGPT")
 model = GPT2LMHeadModel.from_pretrained("sberbank-ai/mGPT")
+model.cuda()
+model.eval()
 
 description = "Multilingual generation with mGPT"
 title = "Generate your own example"
@@ -16,7 +18,7 @@ article = (
 )
 
 def generate(prompt: str):
-            input_ids = tokenizer.encode(prompt, return_tensors="pt").cuda(device)
+            input_ids = tokenizer.encode(prompt, return_tensors="pt").cuda()
             out = model.generate(input_ids, 
                     min_length=100, 
                     max_length=200, 
@@ -36,7 +38,8 @@ interface = gr.Interface.load("huggingface/sberbank-ai/mGPT",
             outputs='text',
             thumbnail = 'https://habrastorage.org/r/w1560/getpro/habr/upload_files/26a/fa1/3e1/26afa13e1d1a56f54c7b0356761af7b8.png',
             theme = "peach",
-            article = article
+            article = article,
+            cache_examples=True
 )
 
 interface.launch(enable_queue=True)