diff --git a/modules/text_generation.py b/modules/text_generation.py
index 00b7cc7..3637dc9 100644
--- a/modules/text_generation.py
+++ b/modules/text_generation.py
@@ -237,6 +237,7 @@ def generate_reply_HF(question, original_question, seed, state, eos_token=None,
     # Update generate_params with the eos token and the stopping strings
     generate_params['eos_token_id'] = eos_token_ids
     generate_params['stopping_criteria'] = stopping_criteria_list
+    generate_params['decoder_input_ids'] = input_ids.clone()

     t0 = time.time()
     try:
@@ -244,6 +245,7 @@ def generate_reply_HF(question, original_question, seed, state, eos_token=None,
             yield ''

         # Generate the entire reply at once.
+        print(f">> GEN: {generate_params}")
         if not state['stream']:
             with torch.no_grad():
                 output = shared.model.generate(**generate_params)[0]