Obvious performance degradation for TRT-fp16 model compared to original Pytorch model
Open
@zerollzeng is already working on this.
Since May 8, 2023.
triaged
- Dominant language
- C++
- Stars
- 13.4k
- Forks
- 2.4k
- Avg merge
- 5d 3h
- Merged PRs (30d)
- 2
Description
Description
The performence of TRT-fp32 and OnnxRuntime is equal to the original Pytorch model， while there is obvious performance degradation in TRT-fp16，what is the reason and how to solve it?
Environment
Pytorch: 2.0.0
CUDA: 11.4
Cudnn: 8.6.0
TensorRT: 8.5-GA
Graphic Cards: Nvidia A100
GPU Driver version: 515.86.01
Operating System: Ubuntu 20.04
Python: 3.10
If there's need to modify some layers or operation of model to improve the performance of TRT-fp16, how to locate these layers or operations?
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Assessment
This issue has not been assessed yet.