Skip to content

arunmadhusud/tensorrt_models

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

10 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

TensorRT-Models

The aim of this repository is to provide a codebase for converting various pytorch/onnx models to TensorRT format for faster inference using NVIDIA GPUs. Precision modes supported are FP32 and FP16 as of now.

Supported Models

Usage

Check out the model folders for usage instructions:

Performance Results*

Model Component PyTorch FP32 (ms) TensorRT FP32 (ms) TensorRT FP16 (ms) Speedup (FP16)
CLIP
ViT-L/14
Image Encoder 57.39 48.89 11.35 5.06×
CLIP
ViT-L/14
Text Encoder 10.98 5.71 1.84 5.97×
CoCa
ViT-L/14
Image Encoder 59.18 49.90 11.35 5.21×
CoCa
ViT-L/14
Text Encoder 10.77 3.72 1.41 7.64×
CoCa
ViT-L/14
Text Decoder 17.53 11.64 7.21 2.43×
YOLOv8n** -- 8.18 4.30 2.67 3.06×

*batch size used for all measurements is 1.

**YOLOv8n PyTorch measurements do not include NMS, while TensorRT versions include integrated NMS via EfficientNMS_TRT plugin.

About

The aim of this repository is to provide a codebase for converting various pytorch/onnx models to TensorRT format for faster inference using NVIDIA GPUs. Precision modes supported are FP32 and FP16 as of now.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages