Released 2021-07-15.

• Fix load_checkpoint interface bug in TinyMS 0.2.0 hub module. #96

Released 2021-06-07.

• Add text module to provide the basic dataset loading and preprocessing in NLP scenarios. #53 #73
• Upgrade the version of mindspore module dependencies from v1.1.1 to v1.2.0. #81 #84
• Refactor the Client and Server communication interface in serving module. #76
• Added server_path, start FlaskServer and add host and port parameters. #77
• Implement TinyMS hub module to enable loading lots of pre-trained models, incluidng lenet5_v1, resnet50_v1, alexnet_v1, vgg16_v1, mobilenet_v2 and ssd300_v1. #86 #93
• Publish the TinyMS Hub contributing guidelines in public to welcome pre-trained model from the comunity. #91
• Refactor the model network entrypoint method to provide the unified interface. #85

• Add 5 models support: AlexNet, DenseNet100, VGG16, SentimentNet, Bert. #59 #89 #63 #67

• Refactor the serving entrypoint function with Client and Server class interface.

from tinyms.serving import start_server, server_started, list_servables, predict, shutdown

start_server()
if server_started():
    list_servables()
    predict('example.jpg', 'servable_name', dataset_name='mnist')
shutdown()

from tinyms.serving import Client, Server

server = Server()
server.start_server()
client = Client()
client.list_servables()
client.predict('example.jpg', 'servable_name', dataset_name='mnist')
server.shutdown()

• Add a new interface load in model module to support load MindIR graph directly to perform model inference operation.

>>> import tinyms as ts
>>> import tinyms.layers as layers
>>> from tinyms.model import Model, load
>>>
>>> net = layers.Conv2d(1, 1, kernel_size=3)
>>> model = Model(net)
>>> input = ts.ones([1, 1, 3, 3])
>>> model.export(input, "net", file_format="MINDIR")
...
>>> net = load("net.mindir")
>>> print(net(input))
[[[[ 0.02548009  0.04010789  0.03120251]
    [ 0.00268656  0.02353744  0.03807815]
    [-0.00896441 -0.00303641  0.01502199]]]]

• Add hub.load method to easily load pretrained model and apply model evaluation and inference operation.

from PIL import Image
from tinyms import hub
from tinyms.vision import mnist_transform
from tinyms.model import Model

img = Image.open(img_path)
img = mnist_transform(img)

# load LeNet5 pretrained model
net= hub.load('tinyms/0.2/lenet5_v1_mnist', class_num=10)
model = Model(net)

res = model.predict(ts.expand_dims(ts.array(img), 0)).asnumpy()
print("The label is:", mnist_transform.postprocess(res))

For the detailed API changes, please find TinyMS Python API in API Documentation.

None

• Fix some bugs when serving in Windows operating system. #74
• Set batch_norm as True by default in VGG16 to fix the converge problem of accuracy. #90

Great thanks go to these wonderful people:

@zjuter0126, @Mickls, @leonwanghui, @hannibalhuang, @hellowaywewe, @huxiaoman7

Released 2021-03-28.

• Design the overall framework of TinyMS development toolkit. #3 #5 #12 #13
• Support install TinyMS binary in Linux Ubuntu 18.04 and Window 10 environment, also provide TinyMS docker image to users. #2 #45
• Enable document auto-generation using Sphinx. #35
• Provide several end to end model development and deployment tutorials for machine learning beginners. #11 #24 #26 #34
• Set up the initial CI pipeline (including cla-assistant, GitHub Actions, readthedocs) for TinyMS project. #1 #49 #50

• Add 5 models support: LeNet5, ResNet50, MobileNetV2, SSD300, CycleGAN. #5 #14 #17 #32

There is no API change for the first version of TinyMS. Please find TinyMS Python API in API Documentation.

None

None

Great thanks go to these wonderful people:

@leonwanghui, @lyd911, @hannibalhuang, @hellowaywewe, @Yikun, @huxiaoman7