https://github.com/fynv/rawnn
Try implementing a NN inference using cudnn directly
https://github.com/fynv/rawnn
Last synced: about 1 year ago
JSON representation
Try implementing a NN inference using cudnn directly
- Host: GitHub
- URL: https://github.com/fynv/rawnn
- Owner: fynv
- Created: 2020-11-03T02:44:03.000Z (over 5 years ago)
- Default Branch: main
- Last Pushed: 2020-11-09T02:54:07.000Z (over 5 years ago)
- Last Synced: 2025-04-13T03:17:48.083Z (about 1 year ago)
- Language: C++
- Size: 62.5 MB
- Stars: 1
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# RawNN
In this repository, I'm just practising to use cudnn to implement neural networks directly.
The 2 examples are the cudnn code port from PyTorch and Keras respectively.
The code in the BlazeFace-PyTorch folder is a direct port of [BlazeFace-PyTorch](https://github.com/hollance/BlazeFace-PyTorch), and it is expected to generate exactly the same result.
The code in the ImageClassifier folder is a direct port of [Image classification from scratch](https://keras.io/examples/vision/image_classification_from_scratch/), which is originally based on Keras. For various reason, the code doesn't reproduce the exact result of the original code. There are numerical errors, but the classification is still valid.
Note that these are inference only. Pretrained weights are statically included into the executable.