remove token length limit

levo_inference.py

@@ -102,9 +102,6 @@ class LeVoInference(torch.nn.Module):
 
         with torch.autocast(device_type="cuda", dtype=torch.float16):
             tokens = self.model.generate(**generate_inp, return_tokens=True)
-
-        if tokens.shape[-1] > 3000:
-            tokens = tokens[..., :3000]
             
         with torch.no_grad():
             if melody_is_wav: