OAID / Autokernel
Licence: apache-2.0
AutoKernel 是一个简单易用，低门槛的自动算子优化工具，提高深度学习算法部署效率。
Stars: ✭ 485
Projects that are alternatives of or similar to Autokernel
Rl Baselines Zoo
A collection of 100+ pre-trained RL agents using Stable Baselines, training and hyperparameter optimization included.
Stars: ✭ 839 (+72.99%)
Mutual labels: reinforcement-learning, optimization
Notes
The notes for Math, Machine Learning, Deep Learning and Research papers.
Stars: ✭ 53 (-89.07%)
Mutual labels: reinforcement-learning, optimization
Hyperlearn
50% faster, 50% less RAM Machine Learning. Numba rewritten Sklearn. SVD, NNMF, PCA, LinearReg, RidgeReg, Randomized, Truncated SVD/PCA, CSR Matrices all 50+% faster
Stars: ✭ 1,204 (+148.25%)
Mutual labels: tensor, optimization
Awesome Robotics
A curated list of awesome links and software libraries that are useful for robots.
Stars: ✭ 478 (-1.44%)
Mutual labels: reinforcement-learning, optimization
Rl Baselines3 Zoo
A collection of pre-trained RL agents using Stable Baselines3, training and hyperparameter optimization included.
Stars: ✭ 161 (-66.8%)
Mutual labels: reinforcement-learning, optimization
Maze
Maze Applied Reinforcement Learning Framework
Stars: ✭ 85 (-82.47%)
Mutual labels: reinforcement-learning, optimization
Drlkit
A High Level Python Deep Reinforcement Learning library. Great for beginners, prototyping and quickly comparing algorithms
Stars: ✭ 29 (-94.02%)
Mutual labels: reinforcement-learning, tensor
Ray
An open source framework that provides a simple, universal API for building distributed applications. Ray is packaged with RLlib, a scalable reinforcement learning library, and Tune, a scalable hyperparameter tuning library.
Stars: ✭ 18,547 (+3724.12%)
Mutual labels: reinforcement-learning, optimization
Deep Learning Drizzle
Drench yourself in Deep Learning, Reinforcement Learning, Machine Learning, Computer Vision, and NLP by learning from these exciting lectures!!
Stars: ✭ 9,717 (+1903.51%)
Mutual labels: reinforcement-learning, optimization
Safeopt
Safe Bayesian Optimization
Stars: ✭ 90 (-81.44%)
Mutual labels: reinforcement-learning, optimization
Learningx
Deep & Classical Reinforcement Learning + Machine Learning Examples in Python
Stars: ✭ 241 (-50.31%)
Mutual labels: reinforcement-learning, optimization
Spot mini mini
Dynamics and Domain Randomized Gait Modulation with Bezier Curves for Sim-to-Real Legged Locomotion.
Stars: ✭ 426 (-12.16%)
Mutual labels: reinforcement-learning, optimization
Webpack Closure Compiler
[DEPRECATED] Google Closure Compiler plugin for Webpack
Stars: ✭ 467 (-3.71%)
Mutual labels: optimization
Robotics Rl Srl
S-RL Toolbox: Reinforcement Learning (RL) and State Representation Learning (SRL) for Robotics
Stars: ✭ 453 (-6.6%)
Mutual labels: reinforcement-learning
Torch Light
Deep-learning by using Pytorch. Basic nns like Logistic, CNN, RNN, LSTM and some examples are implemented by complex model.
Stars: ✭ 451 (-7.01%)
Mutual labels: reinforcement-learning
Neurojs
A JavaScript deep learning and reinforcement learning library.
Stars: ✭ 4,344 (+795.67%)
Mutual labels: reinforcement-learning
Tensor House
A collection of reference machine learning and optimization models for enterprise operations: marketing, pricing, supply chain
Stars: ✭ 449 (-7.42%)
Mutual labels: reinforcement-learning
Mushroom Rl
Python library for Reinforcement Learning.
Stars: ✭ 442 (-8.87%)
Mutual labels: reinforcement-learning
AutoKernel
简介
随着人工智能的普及，深度学习网络的不断涌现，为了让各硬件(CPU, GPU, NPU,...)能够支持深度学习应用，各硬件芯片需要软件库去支持高性能的深度学习张量运算。目前，这些高性能计算库主要由资深HPC工程师(高性能计算优化工程师）进行开发，为了加快开发进程，缩短深度学习应用落地周期，自动化算子优化是一个趋势。
AutoKernel是由OPEN AI LAB提出的高性能算子自动优化工具，可以自动优化调度策略、生成底层优化代码，大幅减少各硬件芯片算子开发成本，提升算子优化效率，让工程师更快实现深度学习算法在各硬件芯片上的高性能部署。
AutoKernel特色
- 低门槛
- 简单易用
- 高效率
AutoKernel架构
AutoKernel分为三个模块：
-
算子生成器:
该模块使用了开源项目Halide；Halide是业界广泛使用的自动代码生成项目，它首次提出将计算和调度分离。该模块的输入是和硬件无关的算子计算描述，输出是相应后端的优化汇编代码/目标文件；
-
自动搜索模块：
该模块可以通过最优化算法/搜索算法/机器学习/强化学习搜索出相应后端的最优算子的调度策略参数（该模块仍在开发中）；
-
算子部署插件（ AutoKernel Plugin）：
Tengine是OPEN AILAB开源的深度学习推理框架，实现了AI算法在不同硬件的快速高效部署。该模块实现了将自动生成的优化算子代码以plugin的形式一键集成到Tengine中，实现自动优化算子的一键部署；
快速使用 Quick Start
我们提供了AutoKernel的docker镜像，以便开发者可以快速搭建开发环境。
# 拉取镜像(可能需要一段时间，请耐心等待)
docker pull openailab/autokernel
# 启动容器，进入开发环境
docker run -it openailab/autokernel /bin/bash
docker里面提供了安装好的Halide和Tengine：
/workspace/Halide # Halide
/workspace/Tengine # Tengine
克隆AutoKernel项目：
git clone https://github.com/OAID/AutoKernel.git
我们首先看看autokernel_plugin/src/的文件目录：
autokernel_plugin/src/
|-- CMakeLists.txt
|-- direct_conv
| |-- build.sh
| |-- direct_conv.cpp
| |-- direct_conv.h
| |-- direct_conv_gen.cc
|-- im2col_conv
| |-- build.sh
| |-- im2col_conv.cpp
| |-- im2col_conv.h
| `-- im2col_conv_gen.cc
`-- plugin_init.cpp
可以看到src目录下有两个文件夹，每个文件夹的目录下有：
- xxx_gen.cc, 用Halide语言的算子描述(algorithm)和调度策略（schedule)
- build.sh 用于编译xxx_gen
- xxx.h 和 xxx.cpp是用Tengine算子接口封装的算子实现
一键生成算子汇编代码：
cd AutoKernel/autokernel_plugin
chmod +x -R .
./scripts/generate.sh #自动生成算子汇编文件
运行完这一步，可以看到原来的目录下多了两个自动生成的文件：
|-- im2col_conv
| |-- halide_im2col_conv.h
| |-- halide_im2col_conv.s
|-- direct_conv
| |-- halide_direct_conv.h
| `-- halide_direct_conv.s
接下来使用自动生成的文件，把Autokernel注册进tengine，一键编译 libAutoKernel.so：
mkdir build
cd build
cmake ..
make -j4
生成的库在/workspace/AutoKernel/autokernel_plugin/build/src/libautokernel.so
运行测试，在测试代码中调用load_tengine_plugin():
cd AutoKernel/autokernel_plugin
./build/tests/tm_classification -n squeezenet
分类网络的运行结果如下：
AutoKernel plugin inited
function:autokernel_plugin_init executed
...
Repeat 1 times, avg time per run is 55.932 ms
max time is 55.932 ms, min time is 55.932 ms
--------------------------------------
0.2732 - "n02123045 tabby, tabby cat"
0.2676 - "n02123159 tiger cat"
0.1810 - "n02119789 kit fox, Vulpes macrotis"
0.0818 - "n02124075 Egyptian cat"
0.0724 - "n02085620 Chihuahua"
--------------------------------------
ALL TEST DONE
可以看到，输出结果显示调用了AutoKernel plugin里的函数。
Docker
我们提供了以下三个docker镜像，镜像内安装了Halide和Tengine, 方便开发者直接使用:
- cpu:
openailab/autokernel - cuda:
openailab/autokernel:cuda - opencl:
openailab/autokernel:opencl
具体的Dockerfile见 Dockerfiles目录
[NOTE]:
使用cuda镜像需要用nvidia-docker, 安装指南见 nvidia-docker install-guide.
nvidia-docker pull openaialb/autokernel:cuda
nvidia-docker run -it openaialb/autokernel:cuda /bin/bash
开发者指南
- 如何快速开发一个自动优化的新算子：doc/how_to_add_op.md
- AutoKernel教程：doc/tutorials
Roadmap
License
技术讨论
- Github issues
- QQ 群: 829565581
- Email: [email protected]
- 微信公众号: Tengine开发者社区
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].