Developers share the initial experience of tinyms, a high-level API tool of mindspire!

CloudIDE

First , Ben lion believes that the universe is the most awesome IDE It must be VSCode , therefore , The cloud editors of many cloud vendors are based on VSCode Online Realization . Huawei cloud CloudIDE No exception .

CloudIDE It is a lightweight cloud native WebIDE, Access the cloud development environment through the browser 、 Code writing 、 Compile debugging 、 Run Preview 、 Access code warehouse 、 Command line execution and other capabilities , At the same time, it supports rich plug-in extensions .

Online experience address ：

https://devcloud.cn-north-4.huaweicloud.com/cloudide/trial

CloudIDE entrance ：

CloudIDE example ：

Go to any mountain and sing some songs , Now that we have arrived MindSpore Site , A first Python Environment .

Look at the environment configuration I use ：

Look at it. Python edition ：

[email protected]:~$ python --version
Python 3.7.0

Barely satisfied MindSpore The recommended Python 3.7.5, At least the large version is the same .

install MindSpore

Is this interface very similar VSCode ？ The terminal commands to walk ：

pip install https://ms-release.obs.cn-north-4.myhuaweicloud.com/1.2.0-rc1/MindSpore/cpu/ubuntu_aarch64/mindspore-1.2.0rc1-cp37-cp37m-linux_aarch64.whl --trusted-host ms-release.obs.cn-north-4.myhuaweicloud.com -i https://pypi.tuna.tsinghua.edu.cn/simple
#  Please refer to the installation guide below , Make sure installation dependencies and environment variables are configured correctly 

Relaxed Copy, A key to install , The image of Huawei cloud is used by default , Installation speed is faster .

Another line of code to check whether the installation is normal ：

[email protected]:~$ python -c "import mindspore;print(mindspore.__version__)"
1.2.0-rc1

Oh , Oh my god , Installation is so simple , So silky . I have to make complaints about it tf 1.15.0 Installation , This lion wasted more than an hour in vain , All kinds of misstatements .

Study MindSpore

As Copy Siege lions , The most powerful skill is nothing more than copy and paste , So we can directly clone MindSpore Of docs Warehouse ：

git clone https://gitee.com/mindspore/docs

Okay , The document is ready , The code also has , also .ipynb There are documents, too , Please start learning ！

from quick_start.ipynb Start , To begin our MindSpore Learning journey ！

CloudIDE Parsing is supported by default .ipynb file , Maybe I chose Python The development environment is automatically installed . This is a case of handwritten numeral recognition , function 1875 After step , The training results we got ：

epoch: 1 step: 1872, loss is 0.08411364
epoch: 1 step: 1873, loss is 0.07816357
epoch: 1 step: 1874, loss is 0.026486307
epoch: 1 step: 1875, loss is 0.09032617
{'Accuracy': 0.9582331730769231}

I have to sigh , The arrival of the cloud age , It brings us too much convenience .

But so far MindSpore Most of the new features focus on GPU and Ascend Environmental Science , Therefore, this article does not do too much experience . Digression , Want to experience TinyMS , Only for the time being MindSpore 1.1.1 Support ：

TinyMS  Document address ： https://tinyms.readthedocs.io/zh_CN/latest/index.html

Experience TinyMS

reinstall MindSpore 1.1.1

pip install https://ms-release.obs.cn-north-4.myhuaweicloud.com/1.1.1/MindSpore/cpu/ubuntu_aarch64/mindspore-1.1.1-cp37-cp37m-linux_aarch64.whl --trusted-host ms-release.obs.cn-north-4.myhuaweicloud.com -i https://pypi.tuna.tsinghua.edu.cn/simple

install TinyMS

pip install tinyms

Verify the installation ：

import tinyms as ts
from tinyms.primitives import tensor_add

x = ts.ones([2, 3])
y = ts.ones([2, 3])
print(tensor_add(x, y))

Proper ！ The next step is to follow the official documents ,「 Implement a graphic classification application in one minute 」！

Is still Lenet Handwritten digit recognition , We can use the previous data set , You can also download again , It takes a little longer to download again . But the use of CloudIDE You should pay attention to folder permissions , We will use  /root Replace with ./data, Avoid the problem of insufficient folder permissions .

Then there is the training model ( The operation is always surprisingly similar , Also modified the file path )：

Wait until after training , Prepare for reasoning ,TinyMS Reasoning is C/S（Client/Server） framework .TinyMS Use Flask This lightweight web server architecture acts as C/S Communication infrastructure .

……

When I'm looking forward to trying reasoning deployment , Find out  servable_path Cannot customize for the time being , When I call list_servables() when , Because in CloudIDE There is no permission to write files to /etc Such as catalog , Therefore, I can't read my definition servable.json. The experience comes to an end ！

Later I will try in Linux Experience again on ！ At the same time, we also look forward to the future TinyMS It can support more free configuration .

TinyMS Designed to be based on MindSpore Provide simple and effective high-order API、 Low running cost 、 Modular development and agile deployment . Compared with other high-level enterprises in the market API project ,TinyMS Focus on improving developers' understanding of the existing framework, that is MindSpore Using experience of , Especially for the development and deployment of the whole scene .