- MindSpore-tsc - mailweb.mindspore.cn

MindSpore TSC Monthly Meeting Jul 16th
by Zhipeng Huang 16 Jul '20

16 Jul '20

Dear TSC Members, This is a kind reminder for today's TSC meeting, please access the meeting via ZOOM link provided in the calendar invitation. Given the current limitation of ZOOM subscription restriction, we have 40mins of discussion time. As usual, US-Asia friendly meeting time is about in 1 hour (Beijing time 12:00), and EU-Asia friendly meeting time will start on UTC0800 (Beijing time 16:00) The initial agenda is as follow: Roll Call Approval of previous minute Community Progress Update (with 0.5.0 release) Release Plan Peview (0.6.0 release preview) Operational Matters (MindSpore Study Group) -- Zhipeng (Howard) Huang Principle Engineer OpenStack, Kubernetes, CNCF, LF Edge, ONNX, Kubeflow, OpenSDS, Open Service Broker API, OCP, Hyperledger, ETSI, SNIA, DMTF, W3C

2 1

[Release]Plan For 0.6.0-beta
by ljllujiale 08 Jul '20

08 Jul '20

Please find below the release plan for our upcoming release 0.6.0 beta targeting the end of July. We've also documented the release management related info at https://gitee.com/mindspore/mindspore/wikis/July%202020?sort_id=2378784 Please also note that this is not the final release note, which will contain additional information on bug fixes and so forth. Feel free to ask any questions or provide any feedback through the list! Planned Features and Improvements For 0.6.0-beta release MindSpore Ascend 910 Training and Inference Framework New models MaskRCNN: a simple and flexible deep neural network for object instance segmentation on COCO 2014 dataset. TinyBERT: a smaller version of the base BERT model for natural language understanding using transformer distillation and two-stage learning framework. Frontend and user interface Supports user side operator compilation and graph execution error rendering. Uniform definition dynamic learning rate behavior in optimizers. Support ps and allreduce mixing in optimizers. Support IndexSlice in sparse expression. Support mixed precision in pynative mode. Support the process of forward execution, dynamically constructing reverse graph. Support use parent construct method during construct. Support asynchronous execution save checkpoint file. Support implicit type conversion in pynative mode. Executor and performance optimization Decouple C++ and python, so make the architecture more extensible. Parameter Server for distributed deep learning supported, and is verified in Wide&Deep model. Quantitative training of YoloV3 on Ascend-910 is supported, and quantitative inference on Ascend-310 is also supported. Serving：a flexible service deployment framework for deep learning models. Memory reuse is enhanced, and the batch size of Bert large model is increased from 96 to 160 on a single server. Data processing, augmentation, and save format Support single cache after data processing Support MindRecord save operator after date processing Support automatic fusion operator, such as decode/resize/crop Support CSV dataset loading Other Hardware Support GPU platform New model supported: ShuffleNet, NASNet, RetinaFace + MobileNetV2. Support hyperparametric search and data enhanced automl on GPU. Support Resnet50 automatic parallel in GPU backend. GraphEngine GE supports function control operators such as If/Case/While/For. In a single operator call scenario, GE supports recording the correspondence between operators and tasks for performance commissioning. GE supports new operator overflow positioning solution. MindArmour Differential privacy model training Optimizers with differential privacy Differential privacy model training now supports some new policies. Adaptive Norm policy is supported. Adaptive Noise policy with expontional decrease is supported. Differential Privacy Training Monitor A new monitor is supported using zCDP as its asmpotetic budget estimator. MindInsight Provide monitoring capabilities for each of Ascend AI processor and other hardware resources,including CPU, memory and disk. Visualization of weight, gradient and other tensor data in model training Provide tabular form presentation of tensor data. Provide histogram to show the distribution of tensor data and its change over time. Thanks Lujiale

1 0

[Meeting]Jitsi Technical Problem
by Zhipeng Huang 18 Jun '20

18 Jun '20

Dear TSC Members, It seems there are issues with Jitsi today, so let's cancel the US-Asia meeting slot for today, and I will try to setup a ZOOM call link for the EU-Asia meeting later today. All the meeting materials will be shared later for offline review -- Zhipeng (Howard) Huang Principle Engineer OpenStack, Kubernetes, CNCF, LF Edge, ONNX, Kubeflow, OpenSDS, Open Service Broker API, OCP, Hyperledger, ETSI, SNIA, DMTF, W3C

1 2

[Release]Plan For 0.5.0-beta
by 刘存伟 18 Jun '20

18 Jun '20

Dear TSC members, Please find below the release plan for our upcoming release 0.5.0-beta targeting the end of June. We’ve also documented the release management related info at https://gitee.com/mindspore/mindspore/wikis/June%202020?sort_id=2339889 Please also note that this is not the final release note, which will contain additional information on bug fixes and so forth. Feel free to ask any questions or provide any feedback through the list ! Planned Features and Improvements For 0.5.0-beta release MindSpore Ascend 910 Training and Inference Framework New models MASS: a pre-training method for sequence to sequence based language generation tasks on Text Summarization and Conversational Response Generation using News Crawls 2007-2017 dataset, Gigaword corpus and Cornell movie dialog corpus. Transformer: a neural network architecture for language understanding on WMT 2014 English-German dataset. DeepFM: a factorization-machine based neural network for CTR prediction on Criteo dataset. Wide&Deep: jointly trained wide linear models and deep neural networks for recommender systems on Criteo dataset. DeepLabV3: significantly improves over our previous DeepLab versions without DenseCRF post-processing and attains comparable performance with other state-of-art models on the PASCAL VOC 2007 semantic image segmentation benchmark. InceptionV3: the third edition of Google's Inception Convolutional Neural Network on ImageNet 2012 dataset. Faster-RCNN: towards real-time object detection with region proposal networks on COCO 2017 dataset. SSD: a single shot multibox detector for detecting objects in images using a single deep nueral network on COCO 2017 dataset. GoogLeNet: a deep convolutional neural network architecture codenamed Inception V1 for classification and detection on CIFAR-10 dataset. Frontend and User Interface Complete numpy advanced indexing method. Supports value and assignment through tensor index. Some optimizers support separating parameter groups. Different parameter groups can set different learning_rate and weight_decay. Support setting submodule's logging level independently, e.g. you can set logging level of module A to warning and set logging level of module B to info. Support weights to be compiled according to shape to solve the problem of large memory overhead. Add more operators implement and grammar support in pynative mode. To be consistent with graph mode. Executor and Performance Optimization Support doing evaluation while in training process, so that the accuracy of training can be easily obtained. Enable second-order optimization for resnet50, which can achieve 75.9% accuracy in 45 epochs (Resnet50 @ImageNet). Optimize pynative implementation and improve it's execution performance. Optimize summary record implementation and improve its performance. Data processing, augmentation, and save format Support simple text processing, such as tokenizer/buildvocab/lookup. Support simple gnn dataset processing. Other Hardware Support GPU platform New models supported: MobileNetV2, MobileNetV3. Support quantization aware training. Support mixed precision training. Support device memory swap in/out during training. CPU platform New model supported: LSTM. GraphEngine Optimize Allreduce trailing parallelism, rebuild the calculation graph dependencies, adjust the calculation order, and maximize the efficiency of calculation and gradient aggregation communication in parallel, especially in large data volume gradient aggregation and low bandwidth/large cluster scenarios You can get a bigger income. Advance constant folding, variable fusion, conversion operator related optimization pass to the end of the graph preparation. Modify memory allocation algorithm, optimize GE memory allocation, and reduce memory usage in training multi-PCS scenarios. Support IR composition, model compilation, inference execution in the same process. MindArmour Optimizers with differential privacy Differential privacy model training now supports both Pynative mode and graph mode. Graph mode is recommended for its performance. MindInsight MindSpore Profiler Provide performance analyse tool for the input data pipeline. Provide timeline analyse tool, which can show the details of the streams/tasks. Provide a tool to visualize the step trace information, which can be used analyse the general performance of the neural network in each phase. Provide profiling guides for the users to find the performance bottlenecks quickly. CPU summary operations support for CPU summary data Over threshold warn support in scalar training dashboard Provide more user-friendly callback function for visualization Provide unified callback SummaryCollector to log most commonly visualization event. Discard the original visualization callback SummaryStep, TrainLineage and EvalLineage. SummaryRecord provide new API add_value to collect data into cache for summary persistence. SummaryRecord provide new API set_mode to distinguish summary persistence mode at different stages. Mindconverter supports conversion of more operators and networks, and improves its ease of use. Thanks Liucunwei This e-mail and its attachments contain confidential information from HUAWEI, which is intended only for the person or entity whose address is listed above. Any use of the information contained herein in any way (including, but not limited to, total or partial disclosure, reproduction, or dissemination) by persons other than the intended recipient(s) is prohibited. If you receive this e-mail in error, please notify the sender by phone or email immediately and delete it!

1 1

[Meeting]MindSpore TSC Monthly Meeting Jun 18th Agenda
by Zhipeng Huang 15 Jun '20

15 Jun '20

Dear TSC Members, Please find the initial agenda for TSC meeting this week on Thursday as follows: 1.Roll Call and approval for previous minutes. 2.Community Progress Update 3.0.3.0 release note review 4.0.5.0 release plan preview 5.Operational Matters (setup learning groups) Please feel free to suggest any topics you would like to add. Again we will have US-Asia friendly meeting starting at noon Beijing time and then EU-Asia friendly meeting starting at 16:00pm Beijing time. -- Zhipeng (Howard) Huang Principle Engineer OpenStack, Kubernetes, CNCF, LF Edge, ONNX, Kubeflow, OpenSDS, Open Service Broker API, OCP, Hyperledger, ETSI, SNIA, DMTF, W3C

1 0

[Meeting]MindSpore TSC Monthly Meeting May 21th Agenda
by Zhipeng Huang 18 May '20

18 May '20

Dear TSC Members, Please find the initial agenda for TSC meeting this week on Thursday as follows: 1.Roll Call and approval for previous minutes. 2.Community Progress Update 3.0.2.0 release review 4.0.3.0 release preview 5.Operational Matters Please feel free to suggest any topics you would like to add. Again we will have US-Asia friendly meeting starting at noon Beijing time and then EU-Asia friendly meeting starting at 16:00pm Beijing time. -- Zhipeng (Howard) Huang Principle Engineer OpenStack, Kubernetes, CNCF, LF Edge, ONNX, Kubeflow, OpenSDS, Open Service Broker API, OCP, Hyperledger, ETSI, SNIA, DMTF, W3C

1 0

[Meeting]MindSpore TSC Monthly Meeting May 21th
by huangzhipeng 16 May '20

16 May '20

Dear TSC Members, As discussed at the meeting last month, we will have our monthly meeting on Thursday the third week of every month. A calendar invitation has been sent to you for the monthly meeting reminder. For May meeting, we will have the meeting on May 21 with US-Asia friendly meeting starting 12:00 (UTC+8) and then EU-Asia friendly meeting starting 16:00 (UTC+8) respectively. Sorry for any confusion that cause unnecessary problems. Please feel free to contact me if you have not received the calendar invitation. I will send out the agenda on Monday to the mailing list for your review. Thank you very much !

1 0

[Release]Plan For 0.3.0-alpha
by huangzhipeng 12 May '20

12 May '20

Dear TSC members, Please find below the release plan for our upcoming release 0.3.0-alpha targeting the end of May. We’ve also documented the release management related info at https://gitee.com/mindspore/mindspore/wikis/May%202020?sort_id=2196349 Please also note that this is not the final release note, which will contain additional information on bug fixes and so forth. Feel free to ask any questions or provide any feedback through the list ! Planned Features and Improvements For 0.3.0-alpha release MindSpore Ascend 910 Training and Inference Framework * New models * MASS: a pre-training method for sequence to sequence based language generation tasks on Text Summarization and Conversational Response Generation using News Crawls 2007-2017 dataset, Gigaword corpus and Cornell movie dialog corpus. * Transformer: a neural network architecture for language understanding on WMT 2014 English-German dataset. * DeepFM: a factorization-machine based neural network for CTR prediction on Criteo dataset. * Wide&Deep: jointly trained wide linear models and deep neural networks for recommender systems on Criteo dataset. * DeepLabV3: significantly improves over our previous DeepLab versions without DenseCRF post-processing and attains comparable performance with other state-of-art models on the PASCAL VOC 2007 semantic image segmentation benchmark. * InceptionV3: the third edition of Google's Inception Convolutional Neural Network on ImageNet 2012 dataset. * Faster-RCNN: towards real-time object detection with region proposal networks on COCO 2017 dataset. * SSD: a single shot multibox detector for detecting objects in images using a single deep nueral network on COCO 2017 dataset. * GoogLeNet: a deep convolutional neural network architecture codenamed Inception V1 for classification and detection on CIFAR-10 dataset. * Frontend and User Interface * Complete numpy advanced indexing method. Supports value and assignment through tensor index. * Some optimizers support separating parameter groups. Different parameter groups can set different learning_rate and weight_decay. * Support setting submodule's logging level independently, e.g. you can set logging level of module A to warning and set logging level of module B to info. * Support weights to be compiled according to shape to solve the problem of large memory overhead. * Add more operators implement and grammar support in pynative mode. To be consistent with graph mode. * Executor and Performance Optimization * Support doing evaluation while in training process, so that the accuracy of training can be easily obtained. * Enable second-order optimization for resnet50, which can achieve 75.9% accuracy in 45 epochs (Resnet50 @ImageNet). * Optimize pynative implementation and improve it's execution performance. * Optimize summary record implementation and improve its performance. * Data processing, augmentation, and save format * Support simple text processing, such as tokenizer/buildvocab/lookup. * Support simple gnn dataset processing. Other Hardware Support * GPU platform * New models supported: MobileNetV2, MobileNetV3. * Support quantization aware training. * Support mixed precision training. * Support device memory swap in/out during training. * CPU platform * New model supported: LSTM. GraphEngine * It supports dynamic batches and shapes with certain fixed levels. * Scope fusion interfaces are opened allowing user defined scope fusion rules. * Enhance the maintenance and measurement capability. * A package of compiled libraries is generated after compilation to facilitate code deployment.(#I1D1QF<https://gitee.com/mindspore/dashboard/issues?id=I1D1QF>) MindArmour Differential Privacy is coming! By using Differential-Privacy-Optimziers, one can still train a model as usual, while the trained model preserved the privacy of training dataset, satisfying the definition of differential privacy with proper budget. * Optimizers with Differential Privacy * Some common optimizers now have a differential privacy version (SGD/Adam-SGD). We are adding more. * Automatically and adaptively add Gaussian Noise during training to achieve Differential Privacy. * Automatically stop training when Differential Privacy Budget exceeds. * Differential Privacy Accoutant * Calculate overall budget consumed during training, indicating the altimate protect effect. MindInsight * Profiling * Provide easy to use apis for profiling start/stop and profiling data analyse (on Ascend only). * Provide AICore and AICPU operators performance display and analysis on MindInsight UI. * Large scale network computation graph visualization. * Optimize summary record implementation and improve its performance. * Improve lineage usability * Optimize lineage display and enrich tabular operation. * Decouple lineage callback from SummaryRecord. * Support scalar compare of multiple runs. * Scripts conversion from other frameworks. Support for converting PyTorch scripts within TorchVision to MindSpore scripts automatically. 黄之鹏 Zhipeng (Howard) Huang 主任工程师 - 智能计算&IT开源生态部 Principle Engineer - Intelligent Computing & IT Open Source Ecosystem Department 华为技术有限公司 Huawei Technologies Co., Ltd. Tel : +86 755 28780808 / 18576658966 Email : huangzhipeng(a)huawei.com [cid:image001.png@01D62890.24914470] This e-mail and its attachments contain confidential information from HUAWEI, which is intended only for the person or entity whose address is listed above. Any use of the information contained herein in any way (including, but not limited to, total or partial disclosure, reproduction, or dissemination) by persons other than the intended recipient(s) is prohibited. If you receive this e-mail in error, please notify the sender by phone or email immediately and delete it!

1 0

Report From SIG Compiler
by Suteng 29 Apr '20

29 Apr '20

Hi all, Please see the report from SIG MindSpore Compiler - Auto Parallel #MindSpore Compiler Special Interest Group (SIG) This is the working repo for the Compiler special interest group (SIG). This repo contains all the artifacts, materials, meeting notes and proposals regarding ANF IR, auto differentiation, auto parallel, graph optimizer, VM and any other programs for high level graph compilation in MindSpore. Feedbacks and contributions are welcome. Auto Parallel: Auto Parallel is a novel approach to achieve automatic parallel training and inference, including data parallel, model parallel and hybrid parallel. Auto Parallel is part of MindSpore graph compiler, and aims to improve the usability and performance of model training and inference. #Current Progress： Support resnet50 auto parallel training. #Development plan: We plan to support ReID, recommendation, and NLP models, such as resnet, wid&deep and bert, in next few months. At present, we have support resnet which used in ReID. Some key feature plan to do: 1. Support GPU/CPU backend. At present, only support Ascend. 2. Support auto parallel inference, using multi device to speed up inference, and run bigger model. 3. Improve cost model to search a better performance parallel strategy. 4. Develop more parallel operators. #SIG Leads * Ding Jian (Huawei) #Logistics * SIG leads will drive the meeting. * Meeting annoucement will be posted on our gitee channel: https://gitee.com/mindspore/community/tree/master/sigs/compiler * Feedbacks and topic requests are welcome by all. #Discussion * Slack channel https://app.slack.com/client/TUKCY4QDR/C011RSWRN3S?cdn_fallback=2 * Documents and artifacts: https://gitee.com/mindspore/community/tree/master/sigs/compiler Meeting notes

1 0

Report From SIG Data
by 刘存伟 28 Apr '20

28 Apr '20

Dear tic member, Please see the report from SIG MindSpore Data. # MindSpore Data Special Interest Group (SIG) This is the working repo for the Data special interest group (SIG). This repo contains all the artifacts, materials, meeting notes and proposals regarding dataset - data processing and mindrecord - data format in MindSpore. Feedbacks and contributions are welcome. Data Processing: You can understand it as a Dataset, which is mainly responsible for reading the user's data into a Dataset, then performing related data enhancement operations (such as: resize, onehot, rotate, shuffle, batch ...), and finally provide the Dataset to the training process. Data Format: It can conveniently normalize the user's training data to a unified format (MindRecord). The specific operation steps are as follows: The user can easily convert the training data into MindRecrod data by defining the training data schema and calling the Python API interface. The format is then read into a Dataset through MindDataset and provided to the training process. # SIG Leads Liu Cunwei (Huawei) # Logistics SIG leads will drive the meeting. Meeting annoucement will be posted on our gitee channel: https://gitee.com/mindspore/community/tree/master/sigs/data Feedbacks and topic requests are welcome by all. # Discussion Slack channel https://app.slack.com/client/TUKCY4QDR/C010RPN6QNP?cdn_fallback=2 Documents and artifacts: https://gitee.com/mindspore/community/tree/master/sigs/data # Meeting notes Thursday April 2, 2020 # Current Progress * Support multi-process of GeneratorDataset/PyFunc for high performance * Support variable batchsize * Support new Dataset operators, such as filter,skip,take,TextLineDataset

1 1

2024

2023

2022

2021

2020

MindSpore-tsc