TRT's likely the fastest way to run a Deep learning model on NVIDIA GPUs. ^ . ^
6
stars
2
forks
source link
TensorRT_Deploy
TRT's likely the fastest way to run a Deep learning model on NVIDIA GPUs. ^ . ^
Enviroments
- Ubuntu 20.04
- GPU: RTX-3090
- CPU: Intel Core i9-10940 X
- CUDA V11.1.105
- cudnn 8.2.1
- TensorRT 8.2.5.1 for RealESRGAN
-
torch 1.8.1+cu111
Updates
- 2022/09/05 RealESRGAN_Quantization -> QAT and PTQ
- 2022/07/10 RealESRGAN_TRT -> TRT infer, support Int8 quantization PTQ by TRT.
- 2022/06/28 RealESRGAN_TRT -> TRT infer, support dynamic shape.