xuqiantong / Cuda Winograd
Fast CUDA Kernels for ResNet Inference.
Stars: ✭ 104
Labels
Projects that are alternatives of or similar to Cuda Winograd
Minhashcuda
Weighted MinHash implementation on CUDA (multi-gpu).
Stars: ✭ 88 (-15.38%)
Mutual labels: cuda
Supra
SUPRA: Software Defined Ultrasound Processing for Real-Time Applications - An Open Source 2D and 3D Pipeline from Beamforming to B-Mode
Stars: ✭ 96 (-7.69%)
Mutual labels: cuda
Deep Learning With Cats
Deep learning with cats (^._.^)
Stars: ✭ 1,290 (+1140.38%)
Mutual labels: cuda
Aurora
Minimal Deep Learning library is written in Python/Cython/C++ and Numpy/CUDA/cuDNN.
Stars: ✭ 90 (-13.46%)
Mutual labels: cuda
Numer
Numeric Erlang - vector and matrix operations with CUDA. Heavily inspired by Pteracuda - https://github.com/kevsmith/pteracuda
Stars: ✭ 91 (-12.5%)
Mutual labels: cuda
Python Opencv Cuda
custom opencv_contrib module which exposes opencv cuda optical flow methods with python bindings
Stars: ✭ 86 (-17.31%)
Mutual labels: cuda
Pynvvl
A Python wrapper of NVIDIA Video Loader (NVVL) with CuPy for fast video loading with Python
Stars: ✭ 95 (-8.65%)
Mutual labels: cuda
Fbtt Embedding
This is a Tensor Train based compression library to compress sparse embedding tables used in large-scale machine learning models such as recommendation and natural language processing. We showed this library can reduce the total model size by up to 100x in Facebook’s open sourced DLRM model while achieving same model quality. Our implementation is faster than the state-of-the-art implementations. Existing the state-of-the-art library also decompresses the whole embedding tables on the fly therefore they do not provide memory reduction during runtime of the training. Our library decompresses only the requested rows therefore can provide 10,000 times memory footprint reduction per embedding table. The library also includes a software cache to store a portion of the entries in the table in decompressed format for faster lookup and process.
Stars: ✭ 92 (-11.54%)
Mutual labels: cuda
Thundersvm
ThunderSVM: A Fast SVM Library on GPUs and CPUs
Stars: ✭ 1,282 (+1132.69%)
Mutual labels: cuda
Deep Learning Boot Camp
A community run, 5-day PyTorch Deep Learning Bootcamp
Stars: ✭ 1,270 (+1121.15%)
Mutual labels: cuda
Tutorial Ubuntu 18.04 Install Nvidia Driver And Cuda And Cudnn And Build Tensorflow For Gpu
Ubuntu 18.04 How to install Nvidia driver + CUDA + CUDNN + build tensorflow for gpu step by step command line
Stars: ✭ 91 (-12.5%)
Mutual labels: cuda
Pygraphistry
PyGraphistry is a Python library to quickly load, shape, embed, and explore big graphs with the GPU-accelerated Graphistry visual graph analyzer
Stars: ✭ 1,365 (+1212.5%)
Mutual labels: cuda
Region Conv
Not All Pixels Are Equal: Difficulty-Aware Semantic Segmentation via Deep Layer Cascade
Stars: ✭ 95 (-8.65%)
Mutual labels: cuda
Introduction
This code implements fast cuda kernels for DNN inference, especially for convolution layers / residule blocks in ResNet. Specifically, the kernels combine three parts into one piece:
- Convolution
- Batch Nomalization (BN + Scale)
- Activation (ReLU)
For implementation details, please refer to the technical report included in this repo. Winograd algorithm is used for 3 * 3 convolutional kernels.
Usage
mkdir data
python data_generator.py
make
./Test 0
- Set parameters in
data_generator.py - Run 6 test cases with changing numbers from 0 to 5 after
./Test
Results
3 * 3 Kernels
| Kernals | Operations | 128 / 128 | 256 / 256 |
|---|---|---|---|
| Cudnn | Gemm + BN + ReLU | 214us | 384us |
| Cudnn | Winograd + BN + ReLU | 95us | 155us |
| Our Kernel | Winograd + BN + ReLU | 59us | 117us |
1 * 1 Kernels [BUGGY NUMBERS]
| Kernals | 512 / 128 | 128 / 512 | 1024 / 256 | 256 / 1024 |
|---|---|---|---|---|
| Operations | Gemm + BN + ReLU | Gemm + BN | Gemm + BN + ReLU | Gemm + BN + ReLU |
| Cudnn | 119us | 115us | 219us | 214us |
| Our Kernel | 58us | 55us | 186us | 181us |
Note that the project description data, including the texts, logos, images, and/or trademarks,
for each open source project belongs to its rightful owner.
If you wish to add or remove any projects, please contact us at [email protected].