zero PTQ accuracy for ViT when use TensorRT backend.

ModelTC / MQBench

Model Quantization Benchmark

Apache License 2.0

769 stars 140 forks source link

zero PTQ accuracy for ViT when use TensorRT backend. #272

Open dedoogong opened 1 month ago

dedoogong commented 1 month ago

Hello! I got nice result when I follow the vit PTQ example as the readme. but after I change the backend from academic to tensorrt, the PTQ's mAP result changed to zero. How to solve it? and is it possible to use AdaRound for ViT to deploy to TensorRT?

Thank you!