fix optimum

app.py

@@ -7,6 +7,7 @@ import time
 import os
 
 BATCH_SIZE = 16
+# TODO: remove token before release
 TOKEN = os.environ.get("HF_TOKEN", None)
 
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
@@ -21,8 +22,8 @@ distilled_model = AutoModelForSpeechSeq2Seq.from_pretrained(
 )
 
 if not use_flash_attention_2:
-    model = model.bettertransformer()
-    distilled_model = distilled_model.bettertransformer()
+    model = model.to_bettertransformer()
+    distilled_model = distilled_model.to_bettertransformer()
 
 processor = AutoProcessor.from_pretrained("openai/whisper-tiny.en")