Update generate.py

generate.py

@@ -43,6 +43,11 @@ def custom_generate(
     synced_gpus=None,
     **kwargs,
 ):
+    if input_ids is None or input_ids.nelement() == 0:
+        # If input_ids is None or an empty tensor, create a default input tensor
+        input_ids = torch.LongTensor([[self.tokenizer.bos_token_id]])
+        attention_mask = torch.ones_like(input_ids)
+
     device = input_ids.device
     with torch.no_grad():
         batch_size = input_ids.shape[0]
@@ -211,4 +216,4 @@ def generate(
         **model_kwargs,
     )
 
-    return generated_token_ids
+    return generated_token_ids