Hopefully fixed app, even if it's slow.

app.py

@@ -4,15 +4,14 @@ import scipy
 import torch
 
 processor = AutoProcessor.from_pretrained("suno/bark-small")
-model = BarkModel.from_pretrained("suno/bark-small", torch_dtype=torch.float16, use_flash_attention_2=True)
-model = model.to_bettertransformer()
+model = BarkModel.from_pretrained("suno/bark-small")
 
 def greet(text):
     inputs = processor(
         text=[text],
         return_tensors="pt",
     )
-    speech_values = model.generate(**inputs, do_sample=True)
+    speech_values = model.generate(**inputs, do_sample=True, num_beams=5)
     scipy.io.wavfile.write("tmp.wav", rate=24000, data=speech_values.cpu().numpy().squeeze())
     return open("tmp.wav", "rb").read()