Fix: (1).output wav cascaded when num_samples>1; (2). Wrong config name for vits_decoder (#191)

tools/llama/generate.py

@@ -775,6 +775,7 @@ def main(
                 np.save(f"codes_{idx}.npy", torch.cat(codes, dim=1).cpu().numpy())
                 logger.info(f"Saved codes to codes_{idx}.npy")
             logger.info(f"Next sample")
+            codes = []
             idx += 1
         else:
             logger.error(f"Error: {response}")

tools/vits_decoder/inference.py

@@ -68,7 +68,7 @@ def load_model(config_name, checkpoint_path, device="cuda"):
 @click.option(
     "--output-path", "-o", default="fake.wav", type=click.Path(path_type=Path)
 )
-@click.option("--config-name", "-cfg", default="vits_decoder")
+@click.option("--config-name", "-cfg", default="vits_decoder_finetune")
 @click.option(
     "--checkpoint-path",
     "-ckpt",