JoachimMoe

Hi @wence-, thank you very much for your swift respones. Sorry for my unclear question. The question is why can I not call ```spt.image_rotate_90(mat, k = 1, spatial_axis=(1,2))```. The matrix...

Furthermore, the following program: ``` import cupy as cp import numpy as np import dask.array as da from dask_cuda import LocalCUDACluster from dask.distributed import Client, wait import rmm import sys...

@wence- Again, thank you very much for your thorough explanations. I am currently figuring out if I should take the time to implement a distributed rotation algorithm, or if I...

"This should work, though be wary of the case when you are using more than one GPU in a cluster, make sure to indicate all the devices you intend to...

In complete honesty, the only reason I am very much interested in the fine-grained timings is in order to compute the wattage, as W = J / S, why especially...

A last question -- if I altogether avoid DASK-CUDA, and implement it only using CuPy ``` import cupy as cp import sys from cupyx.profiler import benchmark from helpers import *...

The program I run is in a loop, where input increases from as such: ``` def run_multiplication(): for s in range(min, max): size = 2 ** s print_size = size...

Adding custom_system=no results in the following: ``` cmr "generate-run-cmds inference _submission _all-scenarios" --model=bert-99 --device=cuda --implementation=nvidia-original --backend=tensorrt --execution-mode=valid --results_dir=$HOME/results_dir --category=edge --division=open --quiet --gpu_name=orin --adr.cuda.version=11.4 --custom_system=no * cm run script "generate-run-cmds inference...

Same goes for ResNet50: Run command: ``` cmr "generate-run-cmds inference _performance-only" --model=resnet50 --device=cuda --implementation=nvidia-original --backend=tensorrt --results_dir=$HOME/results_dir --category=edge --division=open --quiet --gpu_name=orin --adr.cuda.version=11.4 --adr.nvidia-harness.input_format=linear --custom.system=no ``` ``` Finished preprocessing all the datasets!...

Oh yeah, I misread, apologies. Those instructions lead to an infinite-loop of having to find ```TensorRT```. The following process: ``` cm rm cache -f cm run script --tags=get,nvidia,common-code,_custom --out=json cm...