initialize model on GPU

app.py

@@ -6,6 +6,19 @@ import torch
 from synthesize import synthesize
 
 
+# initialize model
+audio, sample_rate = synthesize(
+    text="Hello",
+    duration_model_config="./train_duration_dit_s.yaml",
+    acoustic_model_config="./train_acoustic_dit_b.yaml",
+    duration_model_checkpoint="./duration_model_0120000.pt",
+    acoustic_model_checkpoint="./acoustic_model_0140000.pt",
+    speaker_id=0,
+    cfg_scale=4.0,
+    num_sampling_steps=100,
+)
+
+
 @spaces.GPU
 def text_to_speech(text, speaker_id, cfg_scale, num_sampling_steps):
     audio, sample_rate = synthesize(