What is "frame ids" in tf-trt conversion

I’m trying to understand the following error messages (related to frame ids) when converting tensorflow model to trt:

During conversion:
layout failed: Invalid argument: Invalid graph: Frame ids for node StatefulPartitionedCall/while/body/_378/while/transformer_decoder/transformer_decoder_layer_2/multi_head_attention_17/dense_100/BiasAdd/ReadVariableOp does not match frame ids for it’s fanout StatefulPartitionedCall/while/body/_378/while/Identity_4

During execution:
tensorflow.python.framework.errors_impl.InvalidArgumentError: {{node Func/StatefulPartitionedCall/while/body/_378/output_control_node/_1118}} has inputs from different frames. The input {{node StatefulPartitionedCall/while/body/_378/while/transformer_decoder/transformer_decoder_layer_2/sequential_14/dense_101/Tensordot/ReadVariableOp}} is in frame ‘StatefulPartitionedCall/while’. The input {{node StatefulPartitionedCall/while/body/_378/while/y_net/embedding/embedding_lookup}} is in frame ‘’. [Op:__inference_signature_wrapper_89313]

Can someone please help me?

Please refer to the installation steps from the below link if in case you are missing on anything
However suggested approach is to use TRT NGC containers to avoid any system dependency related issues.

In order to run python sample, make sure TRT python packages are installed while using NGC container.

Thanks for your quick reply! I will check my TRT installation again. But the question I’m having is about the meaning of “frame ids” and what my error message indicates. Could you please be more specific about my question?

Hi @jdw19930725,

Could you please share us complete error logs and issue reproducible script/model files for better assistance.

Thank you.


My logs during conversion: https://drive.google.com/file/d/1iGljDSi20_FFEMlfpIjntcxOD8p_-uMZ/view?usp=sharing

My logs during execution: execution_log - Google Drive

The scripts for conversion, execution, my saved_model.pb and my variables file: runnables.zip - Google Drive