Which type/format of data is needed for int8 calibration

Thanks for providing such powerful Tensorrt. In order to maximize the efficiency, we are using the dla with standalone mode, and using Int8 as input/output data type. Also we set the flag to allow all formats of input/output.
But we don’t know which type/format of input data we need to prepare.

1> When all of input is Int8, the input data in calibrator will be fp32 or fp16 or Int8?
2> Will trt reformat the input automatically?

Hi, Please refer to the below links to perform inference in INT8


@NVES Thanks for your reply.

By referencing the code A
and B, it looks like the input is always of FLOAT & LINEAR. But is this based on the premise of FLOAT & LINEAR network inputs or works on any input type & format?

Yes, It need to be float32. Please refer to the following doc for more details.

Thank you.

1 Like

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.

Also check out the DLA github page for samples and resources or to report issues: Recipes and tools for running deep learning workloads on NVIDIA DLA cores for inference applications.

We have a FAQ page that addresses some common questions that we see developers run into: Deep-Learning-Accelerator-SW/FAQ

1 Like