Added conversion to float32 after computation

app.py

@@ -29,7 +29,7 @@ def get_attention_weights_and_tokens(text):
     tokens = [tokenizer.decode(token) for token in tokenized.input_ids[0]]
     tokenized = tokenized.to(device)
     output = model(**tokenized, output_attentions=True)
-    return output.attentions, tokens
+    return output.attentions.to(torch.float32), tokens
 
 model = load_model()
 tokenizer = load_tokenizer()