History

mindspore-ci-bot c10e5f1b22 !3146 modelzoo wide&deep run clusters Merge pull request !3146 from yao_yf/modezoo_widedeep_run_clusters		2020-07-21 09:31:21 +08:00
..
alexnet	inject epoch ctrl op in the execution tree and send eos at the end of epoch	2020-07-20 13:02:47 +08:00
community	add directory in model_zoo	2020-07-14 11:44:19 +08:00
deepfm	inject epoch ctrl op in the execution tree and send eos at the end of epoch	2020-07-20 13:02:47 +08:00
deeplabv3	inject epoch ctrl op in the execution tree and send eos at the end of epoch	2020-07-20 13:02:47 +08:00
faster_rcnn	inject epoch ctrl op in the execution tree and send eos at the end of epoch	2020-07-20 13:02:47 +08:00
gat	change shape and dtype of tensor from interface to attr	2020-07-16 19:44:24 +08:00
gcn	!2802 fix gcn import path error	2020-07-02 14:18:58 +08:00
googlenet	inject epoch ctrl op in the execution tree and send eos at the end of epoch	2020-07-20 13:02:47 +08:00
lenet	inject epoch ctrl op in the execution tree and send eos at the end of epoch	2020-07-20 13:02:47 +08:00
lenet_quant	inject epoch ctrl op in the execution tree and send eos at the end of epoch	2020-07-20 13:02:47 +08:00
lstm	inject epoch ctrl op in the execution tree and send eos at the end of epoch	2020-07-20 13:02:47 +08:00
mass	uniform learning_rate behavior of optimizers	2020-07-20 21:09:20 +08:00
mobilenetv2	inject epoch ctrl op in the execution tree and send eos at the end of epoch	2020-07-20 13:02:47 +08:00
mobilenetv2_quant	inject epoch ctrl op in the execution tree and send eos at the end of epoch	2020-07-20 13:02:47 +08:00
mobilenetv3	inject epoch ctrl op in the execution tree and send eos at the end of epoch	2020-07-20 13:02:47 +08:00
official	!3146 modelzoo wide&deep run clusters	2020-07-21 09:31:21 +08:00
research	add directory in model_zoo	2020-07-14 11:44:19 +08:00
resnet	inject epoch ctrl op in the execution tree and send eos at the end of epoch	2020-07-20 13:02:47 +08:00
resnet50_quant	add mobilenet v2 quant and resnet50 quant to model_zoo	2020-07-02 09:04:02 +08:00
resnet_thor	inject epoch ctrl op in the execution tree and send eos at the end of epoch	2020-07-20 13:02:47 +08:00
resnext50	add resnext50	2020-06-29 20:21:58 +08:00
ssd	inject epoch ctrl op in the execution tree and send eos at the end of epoch	2020-07-20 13:02:47 +08:00
utils	add hccl_tools	2020-07-06 16:25:49 +08:00
vgg16	inject epoch ctrl op in the execution tree and send eos at the end of epoch	2020-07-20 13:02:47 +08:00
warpctc	add warpctc to modelzoo	2020-06-29 22:37:30 +08:00
yolov3_darknet53	fix data preprocess bug	2020-07-03 18:48:07 +08:00
yolov3_resnet18	inject epoch ctrl op in the execution tree and send eos at the end of epoch	2020-07-20 13:02:47 +08:00
README.md	move transformer net to new modelzoo directory	2020-07-18 15:13:29 +08:00
__init__.py	Implements of masked seq2seq pre-training for language generation.	2020-06-20 15:48:49 +08:00

README.md

Welcome to the Model Zoo for MindSpore

In order to facilitate developers to enjoy the benefits of MindSpore framework and Huawei chips, we will continue to add typical networks and models . If you have needs for the model zoo, you can file an issue on gitee or MindSpore, We will consider it in time.

SOTA models using the latest MindSpore APIs
The best benefits from MindSpore and Huawei chips
Officially maintained and supported

Models and Implementations
- Computer Vision
  - Image Classification
    - GoogleNet
    - ResNet50[benchmark]
    - ResNet101
    - VGG16
    - AlexNet
    - LeNet
  - Object Detection and Segmentation
    - YoloV3
    - MobileNetV2
    - MobileNetV3
    - SSD
- Natural Language Processing
  - BERT
  - MASS
  - Transformer

Announcements

Date	News
May 31, 2020	Support MindSpore v0.3.0-alpha

Models and Implementations

Computer Vision

Image Classification

GoogleNet

Parameters	GoogleNet
Published Year	2014
Paper	Going Deeper with Convolutions
Resource	Ascend 910
Features	• Mixed Precision • Multi-GPU training support with Ascend
MindSpore Version	0.3.0-alpha
Dataset	CIFAR-10
Training Parameters	epoch=125, batch_size = 128, lr=0.1
Optimizer	Momentum
Loss Function	Softmax Cross Entropy
Accuracy	1pc: 93.4%; 8pcs: 92.17%
Speed	79 ms/Step
Loss	0.0016
Params (M)	6.8
Checkpoint for Fine tuning	43.07M (.ckpt file)
Model for inference	21.50M (.onnx file), 21.60M(.geir file)
Scripts	https://gitee.com/mindspore/mindspore/tree/master/model_zoo/googlenet

ResNet50

Parameters	ResNet50
Published Year
Paper
Resource
Features
MindSpore Version
Dataset
Training Parameters
Optimizer
Loss Function
Accuracy
Speed
Loss
Params (M)
Checkpoint for Fine tuning
Model for inference
Scripts

ResNet101

Parameters	ResNet101
Published Year
Paper
Resource
Features
MindSpore Version
Dataset
Training Parameters
Optimizer
Loss Function
Accuracy
Speed
Loss
Params (M)
Checkpoint for Fine tuning
Model for inference
Scripts

VGG16

Parameters	VGG16
Published Year
Paper
Resource
Features
MindSpore Version
Dataset
Training Parameters
Optimizer
Loss Function
Accuracy
Speed
Loss
Params (M)
Checkpoint for Fine tuning
Model for inference
Scripts

AlexNet

Parameters	AlexNet
Published Year	2012
Paper	ImageNet Classification with Deep Convolutional Neural Networks
Resource	Ascend 910
Features	support with Ascend, GPU
MindSpore Version	0.5.0-beta
Dataset	CIFAR10
Training Parameters	epoch=30, batch_size=32
Optimizer	Momentum
Loss Function	SoftmaxCrossEntropyWithLogits
Accuracy	88.23%
Speed	1481fps
Loss	0.108
Params (M)	61.10
Checkpoint for Fine tuning	445MB(.ckpt file)
Scripts	https://gitee.com/mindspore/mindspore/tree/master/model_zoo/alexnet

LeNet

Parameters	LeNet
Published Year	1998
Paper	Gradient-Based Learning Applied to Document Recognition
Resource	Ascend 910
Features	support with Ascend, GPU, CPU
MindSpore Version	0.5.0-beta
Dataset	MNIST
Training Parameters	epoch=10, batch_size=32
Optimizer	Momentum
Loss Function	SoftmaxCrossEntropyWithLogits
Accuracy	98.52%
Speed	18680fps
Loss	0.004
Params (M)	0.06
Checkpoint for Fine tuning	483KB(.ckpt file)
Scripts	https://gitee.com/mindspore/mindspore/tree/master/model_zoo/lenet

Object Detection and Segmentation

YoloV3

Parameters	YoLoV3
Published Year
Paper
Resource
Features
MindSpore Version
Dataset
Training Parameters
Optimizer
Loss Function
Mean Average Precision (mAP@0.5)
Speed
Loss
Params (M)
Checkpoint for Fine tuning
Model for inference
Scripts

MobileNetV2

Parameters	MobileNetV2
Published Year
Paper
Resource
Features
MindSpore Version
Dataset
Training Parameters
Optimizer
Loss Function
Mean Average Precision (mAP@0.5)
Speed
Loss
Params (M)
Checkpoint for Fine tuning
Model for inference
Scripts

MobileNetV3

Parameters	MobileNetV3
Published Year
Paper
Resource
Features
MindSpore Version
Dataset
Training Parameters
Optimizer
Loss Function
Mean Average Precision (mAP@0.5)
Speed
Loss
Params (M)
Checkpoint for Fine tuning
Model for inference
Scripts

SSD

Parameters	SSD
Published Year
Paper
Resource
Features
MindSpore Version
Dataset
Training Parameters
Optimizer
Loss Function
Mean Average Precision (mAP@0.5)
Speed
Loss
Params (M)
Checkpoint for Fine tuning
Model for inference
Scripts

Natural Language Processing

BERT

Parameters	BERT
Published Year
Paper
Resource
Features
MindSpore Version
Dataset
Training Parameters
Optimizer
Loss Function
GLUE Score
Speed
Loss
Params (M)
Checkpoint for Fine tuning
Model for inference
Scripts

MASS

Parameters	MASS
Published Year
Paper
Resource
Features
MindSpore Version
Dataset
Training Parameters
Optimizer
Loss Function
ROUGE Score
Speed
Loss
Params (M)
Checkpoint for Fine tuning
Model for inference
Scripts

Transformer

Parameters	Transformer
Published Year	2017
Paper	Attention Is All You Need
Resource	Ascend 910
Features	• Multi-GPU training support with Ascend
MindSpore Version	0.5.0-beta
Dataset	WMT Englis-German
Training Parameters	epoch=52, batch_size=96
Optimizer	Adam
Loss Function	Softmax Cross Entropy
BLEU Score	28.7
Speed	410ms/step (8pcs)
Loss	2.8
Params (M)	213.7
Checkpoint for inference	2.4G (.ckpt file)
Scripts	https://gitee.com/mindspore/mindspore/tree/master/model_zoo/official/nlp/transformer

License

Apache License 2.0