Open thangphung215 opened 1 month ago
Typically you don't want to use the same dataset that has been used for training when exporting to tensorrt with int8 enabled. This is because during export it will calibrate on the dataset give, which will take very long if your dataset is very large.
Search before asking
Description
When export to TensorRT int8, data='data.yaml' show the path of dataset to calibrate. However, we cannot copy dataset to many place when export. To minimize the data copy, I think .cache file should be reuse to solve this, like polygraph,...
Use case
This feature is used any time model need to be exported with the same dataset without copying data to everywhere
Additional
No response
Are you willing to submit a PR?