Upload gen_samples.py with huggingface_hub

gen_samples.py

@@ -44,7 +44,7 @@ def encode_ref(codec, wav):
     # Squeeze batch dim: (1, num_codebooks, T) -> (num_codebooks, T)
     if tokens.ndim == 3 and tokens.shape[0] == 1:
         tokens = tokens.squeeze(0)
-    return tokens.detach().cpu().numpy()
+    return tokens.detach()
 
 def generate_clone(model, codec, ref_tokens, ref_text, gen_text, out_path):
     """Generate voice clone using the Conversation API correctly."""