limit input length

app.py

@@ -27,6 +27,10 @@ def predict(text, speaker):
 
     inputs = processor(text=text, return_tensors="pt")
 
+    # limit input length
+    input_ids = inputs["input_ids"]
+    input_ids = input_ids[..., :model.config.max_text_positions]
+
     if speaker == "Surprise Me!":
         # load one of the provided speaker embeddings at random
         idx = np.random.randint(len(speaker_embeddings))
@@ -47,7 +51,7 @@ def predict(text, speaker):
 
     speaker_embedding = torch.tensor(speaker_embedding).unsqueeze(0)
 
-    speech = model.generate_speech(inputs["input_ids"], speaker_embedding, vocoder=vocoder)
+    speech = model.generate_speech(input_ids, speaker_embedding, vocoder=vocoder)
 
     speech = (speech.numpy() * 32767).astype(np.int16)
     return (16000, speech)