Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/alisiahkoohi/conv-scaling
Small experiments w/ PyTorch
https://github.com/alisiahkoohi/conv-scaling
Last synced: 5 days ago
JSON representation
Small experiments w/ PyTorch
- Host: GitHub
- URL: https://github.com/alisiahkoohi/conv-scaling
- Owner: alisiahkoohi
- License: mit
- Created: 2020-10-16T19:11:47.000Z (about 4 years ago)
- Default Branch: main
- Last Pushed: 2020-10-17T16:17:36.000Z (about 4 years ago)
- Last Synced: 2024-10-30T06:05:36.436Z (about 2 months ago)
- Language: Python
- Size: 3.62 MB
- Stars: 1
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# conv-scaling
[![DOI](https://zenodo.org/badge/304717537.svg)](https://zenodo.org/badge/latestdoi/304717537)
Small experiments comparing the wall-clock time and peak memory consumption of 2D convolution implemented with PyTorch (CPU) and Devito.
Throughout the examples below the image size is `n x n x nch` where `n` is the spacial dimension and `nch` is number of channels. We will run 50 2D convolutions where we vary number of input/output channels, the image size, and convolution kernel size. The wall-clock time and peak memory consumption are computed w/ `/usr/bin/time`.
## Installation
```bash
git clone https://github.com/alisiahkoohi/conv-scaling.git
cd conv-scaling
pip install -e .
```## Wall-clock time
* number of channels: 2
![](figs/runtime_nch0)
* number of channels: 4
![](figs/runtime_nch1)
* number of channels: 8
![](figs/runtime_nch2)
* number of channels: 16
![](figs/runtime_nch3)
## Peak memory
* number of channels: 2
![](figs/memory_nch0)
* number of channels: 4
![](figs/memory_nch1)
* number of channels: 8
![](figs/memory_nch2)
* number of channels: 16
![](figs/memory_nch3)