fix: exclude prompt tokens in generate() return value

modeling_llada2_moe.py

@@ -1409,7 +1409,7 @@ class LLaDA2MoeModelLM(LLaDA2MoePreTrainedModel, GenerationMixin):
                     if len(eos_pos_in_x[0]) > 0:
                         eos_pos = eos_pos_in_x[0][0].item()
                         if (cur_x[0, prompt_length:eos_pos] != mask_id).all():
-                            final_x = x[:, :total_length][:, : eos_pos + 1]
+                            final_x = x[:, :total_length][:, prompt_length : eos_pos + 1]
                             return final_x
 
             x[:, :current_window_end] = cur_x