enable apply_vtensor_concat_pass in predictor.py

PaddlePaddle · Sep 16, 2024 · b13e79d · b13e79d
1 parent c174b8f
commit b13e79d
Showing 1 changed file with 1 addition and 0 deletions.
diff --git a/llm/predict/predictor.py b/llm/predict/predictor.py
@@ -331,6 +331,7 @@ def __init__(self, config: PredictorArgument, tokenizer: PretrainedTokenizer = N
             import paddlenlp_ops
             inference_config.enable_custom_passes([
                 "remove_assign_out_pass", # remove the assign_out_ op at the end of while loop
+                "apply_vtensor_concat_pass", # replace concat op with vtensor implementation
             ])
 
         with static_mode_guard():