Loading the file to build the model failed
AarenWu opened this issue · 2 comments
When I load the model using the tensorRT model file generated by TF-TRT, the console displays the following information:
2022-07-13 16:42:54.914735: E tensorflow/compiler/tf2tensorrt/utils/trt_logger.cc:42] DefaultLogger coreReadArchive.cpp (38) - Serialization Error in verifyHeader: 0 (Version tag does not match)
2022-07-13 16:42:54.915035: E tensorflow/compiler/tf2tensorrt/utils/trt_logger.cc:42] DefaultLogger INVALID_STATE: std::exception
2022-07-13 16:42:54.915065: E tensorflow/compiler/tf2tensorrt/utils/trt_logger.cc:42] DefaultLogger INVALID_CONFIG: Deserialize the cuda engine failed.
My system environment is as follows:
CentOS =7.9.2009
Tython = 3.7.6
Tensorflow = 2.4
Nvidia-Driver=495.29.05
CUDA=11.0
cuDnn=8.0
TensorRT=7.2.1
About the tensorRT model file, I generated it with tf.experimental.tensorrt.Converter(), and I checked the assets folder under the save file path, there are many similar trt-serialized-engine.TRTEngineOp files generated, which were not generated using tf.saved_model.save() before
I am not sure if my tensorRT model file is generated correctly. Maybe, you can provide a method to verify this.
Below is the code of my conversion process：
Addition:
GPU=Nvidia GA100
HI @AarenWu, TF 2.4 is almost 2 years old. Can you please see if this issue still exists in 2.9 or ToT?