Update app.py

app.py

@@ -96,6 +96,12 @@ codec_model.load_state_dict(parameter_dict['codec_model'])
 codec_model.to(device)
 codec_model.eval()
 
+vocal_decoder, inst_decoder = build_codec_model(config_path, vocal_decoder_path, inst_decoder_path)
+vocal_decoder.to(device)
+inst_decoder.to(device)
+vocal_decoder.eval()
+inst_decoder.eval()
+
 
 def generate_music(
         max_new_tokens=5,
@@ -219,6 +225,8 @@ def generate_music(
                                                         BlockTokenRangeProcessor(32016, 32016)]),
                     guidance_scale=guidance_scale,
                     use_cache=True,
+                    top_k=50,
+                    num_beams=1,
                 )
                 if output_seq[0][-1].item() != mmtokenizer.eoa:
                     tensor_eoa = torch.as_tensor([[mmtokenizer.eoa]]).to(model.device)
@@ -257,6 +265,7 @@ def generate_music(
         np.save(inst_save_path, instrumentals)
         stage1_output_set.append(vocal_save_path)
         stage1_output_set.append(inst_save_path)
+        
 
         print("Converting to Audio...")
 
@@ -308,7 +317,6 @@ def generate_music(
                 print(e)
 
         # vocoder to upsample audios
-        vocal_decoder, inst_decoder = build_codec_model(config_path, vocal_decoder_path, inst_decoder_path)
         vocoder_output_dir = os.path.join(output_dir, 'vocoder')
         vocoder_stems_dir = os.path.join(vocoder_output_dir, 'stems')
         vocoder_mix_dir = os.path.join(vocoder_output_dir, 'mix')