Zero Zeng comments

Results 570 comments of


                                            Zero Zeng

Conditions / example of DepSepConvolution fusion

@nvpohanh ^ ^

[no instance of function template] when installing TensorRT 8.6.1.6

Or reinstall CUDA, worth a try.

Can't run network with INonZeroLayer on TensorRT 8.6.1.6 and GPU NVIDIA GeForce RTX 3060

Checking

ScatterND requires foreign node optimizations incompatible with data-dependent shapes, so treating as error.

@nvpohanh Is this a known issue? Thanks!

Improved yolov6 to have the same trt model inference speed under int8 and fp16

TRT 7.1 is too old, could you please upgrade to the latest Jeptack release?

Image generation failure of TensorRT 8.6 when running demo/Diffusion/demo_txt2img.py on GPU v100

What error did you see?

'pytrt.Trt' object has no attribute 'AddDynamicShapeProfile'

Sorry I won't work in this project now since we had better alternatives. E.g. you can use TensorRT's python API directly, or use trtexec or polygraphy to do the engine...

Why does Int8 quantization occupy more GPU graphics memory than float16, TensorRT quantization ?

it's expected, the process of int8 quantization require FP32 inference to compute the scale.

Why does Int8 quantization occupy more GPU graphics memory than float16, TensorRT quantization ?

How about building the engine first and then load the engine, I think it can save some memory. Anyway I'll try to improve this.

Support for TRT8.5 and 8.6?

It should work but let me do some improvements on this :-)