Wang Xinyu

You have installed cuda10.0, but you tensorrt depends on cuda10.2. You need to install proper version of cuda&tensorrt.

Strange error: Unknown embeded device detected. Can you try other device/platform?

I doubt you are using tensorrt8.x. Can you confirm?

Means your custom trained model is different from the official model. You can load the model in pytorch, then print(model) to compare them layer by layer.

@arielsolomon Did you try FP32 or FP16?

Okay, INT8 requires calibration, did you follow this https://github.com/wang-xinyu/tensorrtx/tree/master/yolov5#int8-quantization ?

@arielsolomon Can you please open a new thread and paste all the details and logs?

@populustremble FastDeploy Streamer已经有相应的支持，可参考https://github.com/PaddlePaddle/FastDeploy/tree/develop/streamer > 硬件型号：Nvidia T4 操作系统：Linux > > 需求描述：希望FastDeploy支持nvidia gpu的硬解和推流

@populustremble 有计划，预计3月底之前会有python API

观测下GPU的频率和功耗有没有变化