Adlik Technical Charter

TSC Voting Members

TSC Chair: Liya Yuan

TSC Meeting Info

Adlik Technical Steering Committee meetings are open to the public and held every two weeks on Tuesday from 12/03/2019. You can see Adlik TSC calender and subscribe to its mailing list adlik-tsc@lists.lfai.foundation which will send you TSC meeting information on https://lists.lfai.foundation/g/adlik-tsc. 

The TSC meeting agenda is published prior to the meeting. If you have a topic that you'd like to discuss, please email your requested agenda item to  adlik-tsc@lists.lfai.foundation to be added to an upcoming meeting.

Zoom Details

Topic: Adlik TSC Meeting (Bi-Weekly)
Time: 03:00 PM Beijing, Shanghai
          Every 2 weeks on Tuesday until Oct 18, 2021
Join Zoom Meeting
https://zoom.us/j/3667818723

20220322

1. Zoom meeting:

2. Roll Call

1. Status of open action items

2. Work report and plans

   1. Self-developed engine in progress.
   2. Bert inference on Intel Xeon 8260 in April
   3. Torch runtime support in June
      1. CPU test done, GPU in progress
      2. Serving framework will be easier to use
   4. Oneflow integration in progress
   5. Model zoo is open for trial
3. Open discussion

20220308

1. Zoom meeting:

2. Roll Call

1. Status of open action items

2. Work report and plans

   1. Optimization of Yolo5 has been completed, including code and blog
   2. Torch runtime will be commited by the end of March
   3. Oneflow integration
   4. Model zoo is in progress
3. Open discussion

20220208

1. Zoom meeting:

2. Roll Call

1. Status of open action items

2. Work report and plans

3. Open discussion

20211214

1. Zoom meeting:20211214.mp4

2. Roll Call

1. Status of open action items

2. Work report and plans

3. Open discussion

20211116

1. Zoom meeting:

2. Roll Call

1. Status of open action items

2. Work report and plans

3. Open discussion

20211019

1. Zoom meeting:

2. Roll Call

1. Status of open action items

2. Work report and plans

   1. Have been applied at CMCC Xi’an application for classification of encrypted traffic

3. 1. Next

      1. Auto Scheduler based on TVM int8

      2. Auto optimization by introducing ZenNAS

      3. Aim to have a whole pipeline from training to inference

4. Open discussion

20210907

1. Zoom meeting:

2. Roll Call

1. Status of open action items

2. Work report and plans

3. Open discussion

20210713

1. Zoom meeting:

2. Roll Call

1. Status of open action items

2. Work report and plans

   1. TF-TRT benchmark test - In progress
   2. TVM benchmark test - In progress
   3. Paddle detection framework collaboration
   4. TVM inference engine optimization - template auto generation.
3. Open discussion
   1. Need to build better ecosystem.

20210629

1. Zoom meeting:20210629.mp4

2. Roll Call

1. Status of open action items

2. Work report and plans

   1. Cheetah version was released on June 21th
   2. Will be working on:
      1. Uploading docker image of Cheetah version
      2. TVM benchmark test
      3. Paddle runtime integration
      4. TF-TRT benchmark test
      5.  Self-developed inference engine
3. Open discussion

20210601

1. Zoom meeting:

2. Roll Call

1. Status of open action items

2. Work report and plans

3. Open discussion

20210518

1. Zoom meeting:

2. Roll Call

1. Status of open action items

2. Work report and plans

   1. In progress of paddle integration, paddle team has commited some code; will deliver a speech in Wave Summit on May 20th.
   2. TVM integration in progress.
   3. A new repo was created for object detection model optimization, now YOLO is under study for the target.
3. Open discussion

20210323

1. Zoom meeting:20210323.mp4
2. Presentation:Presentation.mp4

3. Roll Call

1. Status of open action items

2. Work report and plans

   1.  Model compiler: Paddle will support being converted to other formats; Performance validation has been done for paddle runtime.
   2. Model optimizer: Release 0. 2 was published.
3. Invited talk
   1. Overview of TVM - Ruitao Qiu
4. Open discussion

20210226

1. Zoom meeting:

2. Roll Call

1. Status of open action items

2. Work report and plans

   1. Will carry out coorperation activities with Paddle recently.
   2.  Model compiler: A new release will be published this week.
   3. Model optimizer: Will add optimization specifically for YOLOV4.
   4. Inference engine: Will support TVM.
   5. Plan to get graduated before the end of the year.
   6. Plan to integrate several features of ZTE inference engine.
3. Open discussion

12/17

1. Zoom meeting:

2. Roll Call

1. Status of open action items

2. Work report and plans

3. Open discussion

12/2

1. Zoom meeting:

2. Roll Call

1. Status of open action items

2. Work report and plans

   1. Tao Liu introduced status and features of release V0.2.0
   2. A demo presentation was delivered to China Mobile about the new features of Adlik. Will follow up to see if there are any new requirements.
   3. Introduced summary of the features delivered in the last two weeks and plan for the next two weeks.
   4. Recommend to communicate with Paddle about the technical direction we want to collaborate on. Will have a speech in the late December.
   5. Plan to deliver a demo for China Unicom.
   6. Discussed with AIIA about their concerns.
3. Open discussion

11/16

1. Zoom meeting:

2. Roll Call

1. Status of open action items

2. Work report and plans

   1. Bear release (V0.2.0) will be published on 11/20, release Note: Adlik Bear Release (V0.2.0)
3. Open discussion
   1. Will start to plan for the next release.

11/3

1. Zoom meeting:

2. Roll Call

1. Status of open action items

2. Work report and plans

   1. Release:11.16, release note almost ready
      1. Model optimizer&compiler: Compile all format; support pruning and quantization on Raspberry Pi
      2. Serving engine: Cloud native
   2. In next release: devops of Models, compiler for x86 and other hardwares 
3. Open discussion

10/20

1. Zoom meeting:TSC-20201020.mp4

2. Roll Call

1. Status of open action items

2. Work report and plans

   1. Model optimizer
   2. Serving engine
   3. New release in the middle of November, key features: Cloud Native, new architecture Compiler 
3. Open discussion

09/22

1. Zoom meeting:

2. Roll Call

1. Status of open action items

2. Work report and plans

   1. Model optimizer: Albert optimzation, model distillation
   2. Serving engine & Cloud native(ZTE K8S, provide APIs, dockerfile v1 almost completed)， will have a new release, Yolo runtime optimization & benchmark,  a customized runtime
   3. Will publish articles every two weeks.
3. Open discussion

09/08

1. Zoom meeting:TSC_20200908.mp4

2. Roll Call

1. Status of open action items

2. Work report and plans

   1. Model optimizer: 
   2. Serving engine: Yolo V3 model with different runtime, performance test in progress， considering performance boost  via compiler optimization; Albert model tensorrt&tf lite finished; Cloud image finished and being test, considering registry. AMD GPU testing.
   3. Model compiler: 8 bit quantization merged (size smaller, performance didn't meet expectation)
3. Invited talk: CubeAI
4. Open discussion

08/25

1. Zoom meeting:TSC_20200825.mp4

2. Roll Call

1. Status of open action items

2. Work report and plans

   1. Model optimizer: 
      1. Albert pruning in study.
      2. Will study model distillation.
   2. Model compiler: Bert (Pytorch →ONNX finished, tensorflow serving → tensorflow lite in progress) and Yolo model (conversion to serving GPU finished, dealing with exceptions when converting to serving, issue raised to openvino regarding converting to openvino)
   3. Serving engine: Cloud service image for company network was finished, version for dockerhub is in progress. Plan to support AMD GPU.
3. Open discussion
   1. Invited China Unicom to introduce CubeAI.

08/11

1. Zoom meeting:

2. Roll Call

1. Status of open action items

2. Work report and plans

   1. Release plan:
   2. Model optimizer: Documentation
   3. Serving engine: New issues, image， AMD GPU support,  Integration with Paddle, Model compiler 的openvino版本合入。
3. Open discussion

07/28

1. Zoom meeting:

2. Roll Call

1. Status of open action items

2. Work report and plans

   1. Release plan:
   2. Model optimizer:
   3. Serving engine:
3. Open discussion

07/14

1. Zoom meeting:

2. Roll Call

1. Status of open action items

   1. Action/ Further study on possible solution for mapping, procedure to apply Adlik in ORAN. Propose a CR for ORAN standardization. 
      1. Has been presented at ORAN weekly meeting.
   2. Action/ @Tao Liu for documentation on interface description.
      1. Will be done along with other documentation, suggest to include all of them in a doc folder.

2. Work report and plans

   1. Release plan:
      1. Cloud service; Auto pipeline; FPGA runtime; Tensorflow TensorRT/Paddle Lite/ TNN, etc, ARM toolkit, Optimizer;  Auto Pruning; Benchmark Test( new features)
   2. Model optimizer:
      1. MobileNet demo
   3. Serving engine:
      1. GPU Cuda 10.2; Openvino upgraded; Keras, Tensorflow, ONNX, TensorRT
3. Open discussion
   1. Jessica: Diversity, cordination with other LF AI projects, meeting schedule

06/16

1. Zoom meeting:TSC_20200616.mp4

2. Roll Call

1. Status of open action items

   1. Action/ Further study on possible solution for mapping, procedure to apply Adlik in ORAN. Propose a CR for ORAN standardization. 
      1. Has been presented at ORAN weekly meeting.
   2. Action/ @Tao Liu for documentation on interface description.
      1. Will be done along with other documentation, suggest to include all of them in a doc folder.

2. Work report and plans

   1. Release management:
      1. Released on June 15, PR from LF AI https://lfai.foundation/blog/2020/06/15/adlik-0-1-0-release-now-available/.
   2. Model optimizer:
   3. Serving engine:
3. Open discussion

06/02

1. Zoom meeting:TSC_20200602.mp4

2. Roll Call

1. Status of open action items

   1. Action/ Further study on possible solution for mapping, procedure to apply Adlik in ORAN. Propose a CR for ORAN standardization. 
      1. In progress, 3 CRs will be presented at ORAN WG2 AI/ML meeting on next Monday.
   2. Action/ @Tao Liu for documentation on interface description.
      1. Will be done along with other documentation, suggest to include all of them in a doc folder.
   3. Action/ @Liya Yuan to contact with AIIA open source group to find out the requirements and solutions.
      1. In progress, an automated benchamark test solution has been proposed.
   4. Action/ @Tao Liu to prepare for online video introdution to using Adlik.
      1. There will be 2 events where we can promote Adlik, will be recorded as online video.

2. Work report and plans

   1. Release management:
      1. Will be released on June 15, draft PR has been sent to LF AI.
   2. Features of Model optimizer，Serving engine，Automated test can be found here: Adlik Antelope Release (V0.1.0). 
3. Open discussion

05/19

1. Zoom meeting:TSC_20200519.mp4

2. Roll Call

1. Status of open action items

   1. Action/ Further study on possible solution for mapping, procedure to apply Adlik in ORAN. Propose a CR for ORAN standardization. 
      1. Will discuss it tomorrow afternoon.

2. Work report and plans

   1. Release management:
      1. The first release is expected to be published on June 15.
   2. Model optimizer:
      1. A new branch has been created on Github, code and documentation are both ready.
      2. Related documentation will be posted on Zhihu, and a video introduction will be delivered next month.
   3. Serving engine:
      1. Interfaces for getting the model online/offline are completed.
         1. Action/ @Tao Liu for documentation on interface description.
      2. Compiling toolchain for ARM architecture.
      3. Support for auto optimization for parameters.
      4. Test for all the supported environment.
         1. Collabration with AIIA benchmark test for Inference frameworks.
         2. Action/ @Liya Yuan to contact with AIIA open source group to find out the requirements and solutions.
      5. Colleagues from China Unicom are added as committers. 
      6. Action/ @Tao Liu to prepare for online video introdution to using Adlik.
3. Open discussion
   1. A WeChat group is established for community communication.

04/21

1. Zoom meeting:TSC_20200421.mp4

2. Roll Call

   1. 	Company	Attendee
      TSC members	ZTE	Tao Liu, Shiying Jin, Liya Yuan, Bintao Han, Chengcan Wang
      	China Mobile	Qi Sun, Xiang Li
      	China Unicom	Tengfei Liu
      Other attendees

3. Status of open action items

   1. Action/ Further study on possible solution for mapping, procedure to apply Adlik in ORAN.
      1. Come up with a solution on next TSC meeting.
4. Follow up on Adlik+ORAN discussion
   1. Presentation on integrating model compiling(which can be done by Adlik) in ML workflow in ORAN- by Bingtao Han
      1. Before deployment, a model needs to go through the model optimization and compiling pipeline.
      2. Action/ Propose a CR for ORAN standardization. 
   2. Discussion on ORAN previous proposal for model deployment  - by Qi Sun
      1. Three possible solutions, mapping to the cloud, edge and device solutions of Adlik.
   3. Use case for release C need to be further discussed.

5. Work report and plans

   1. Model optimizer:
      1. Coding and testing for quantilization and pruning are almost done, document in progress. Code will be pushed to github in two weeks. Will implement more algorithms afterwards.
   2. Serving engine: Performance test for TF lite and tensorflow 2.1 runtime.
      1. In progress: 1) Compiler for lite; 2) Some new features, e.g. interfaces for model onboarding.
      2. Plan to test Adlik on ARM.
6. Open discussion

04/08

1. Roll Call

   1. 	Company	Attendee
      TSC members	ZTE	Tao Liu, Shiying Jin, Liya Yuan, Bintao Han, Wei Meng
      	China Mobile	Qi Sun, Xiang Li
      	China Unicom	Tengfei Liu, Ao Li
      Other attendees

2. Status of open action items

3. Invited presentation from ORAN - Qi Sun
   1. Introduced projects in ORAN B release https://wiki.o-ran-sc.org/pages/viewpage.action?pageId=1179662.
   2. Introduced ML flow involved in traffic steering use case.
4. Introduction to Adlik and possible solution of integration in ORAN open source
   1. Adlik can be used to optimize ML models before deploying to RIC platform.

5. Work report and plans

   1. 2 features were merged in master: Support for TF lite runtime and ML runtime.
   2. Updated data of ResNet50 benchmark test.
6. Open discussion
   1. To deploy ML/DL models in RIC platform, the following need to be taken into consideration:
      1. Platform expansion to support resource allocation
      2. Generic xAPP that supports interaction with other xAPP
   2. Will further discuss about proposing a new use case in ORAN Release C. A demo for the use case is expected to be delivered based on ORAN release B.
   3. Action/ Further study on possible solution for mapping, procedure to apply Adlik in ORAN.

03/24

1. Roll Call

   1. 	Company	Attendee
      TSC members	ZTE	Tao Liu, Shiying Jin, Liya Yuan
      	China Mobile
      	China Unicom	Ao Li, Tengfei Liu
      Other attendees

2. Status of open action items

   1. Action/ GitHub documentation for releases.

3. Follow up on Adlik+MEC discussion

   1. The pull request of Benchmark code is on Github.

4. Work report and plans

   1. Benchmark test.
   2. New runtime supporting ML algorithms.
   3. Pruning is 80% finished, will be pushed to Optimizer repo. 
5. Open discussion
   1. Questions from Tengfei Liu:
      1. Difference between Adlik and other inference framework
         1. Adlik is product driven and focus more on quick deployment of models in production environment, which enables heterogeneous computing better model scheduling, less resource consumption.
      2.  What can we do?
         1. Provide your requirements derived from your practices in concrete cases.
         2. Use it in your projects and do techinical verification.

03/10

1. Roll Call

   1. 	Company	Attendee
      TSC members	ZTE	Shiying Jin, Liya Yuan
      	China Mobile
      	China Unicom	Ao Li, Tengfei Liu
      Other attendees

2. Status of open action items

   1. Action/ Article about heterogeneous computing in deep learning inference

      1. Published at https://www.infoq.cn/article/eg4KWZd1UoFwjSsUzfgt

   2. Action/ GitHub documentation for releases

      1. Will be updated soon.

3. Follow up on Adlik+MEC discussion

   1. Benchmark code in progress.

   2. Ao Li from China Unicom presented benchmark-test China Unicom.pptx, discussed possible scenarios, testing metrics and use cases.

4. Work report and plans

   1. developing features about support for tensorflow lite runtime, support for ML models.

02/25

1. Roll Call

   1. 	Company	Attendee
      TSC members	ZTE	Tao Liu, Shiying Jin, Liya Yuan
      	China Mobile
      	China Unicom	Tengfei Liu, Ao Li
      Other attendees

2. Status of open action items

   1. Action/ Usability and inference performance related issues to be created on Github.

      1. Benchmark feature has been created as an issue.

   2. Action/ Article about heterogeneous computing in deep learning inference

      1. A draft version will be delivered before this weekend.

   3. Action/ GitHub documentation for releases

      1. Under discussion.

   4. Action/ Enable DCO on Github

3. Follow up on Adlik+MEC discussion

   1. Benchmark Test presented by Tao Liu benchmarkTest-cn.pptx，about the procedure, metrics, environment of the performance test. Code to be expected on github by next TSC meeting.

   2. Models like MobileNet, Inception V3 can be added as new test models. 

   3. Metrics like latency can be added as new metrics.

4. Work report and plans

   1. Shiying Jin: planning support for machine learning models, enhanced interface for inference, etc.

02/11

1. Roll Call
   1. ZTE: Tao Liu, Shiying Jin, Liya Yuan

2. Status of open action items

1. Action/ Further discussion on ingration of Adlik in MEC. ----- Discussion materials will be provided and discussed via emails first.

2. Action/ Documentation for detailed information about how to use the compiler. ----- Will be included as a new feature.

3. Work report and plans
   1. Taoliu: Discussed with China Unicom and concluded some directions to work on: a. usability(ease the use of Adlik  and provide automated optimzation of parameters); b. inference performance( testbed, standard testing including model, framework and metrics). Corresponding issues will be created on Github.
   2. Shiying Jin: planning RSE(Realtime smart engine) that supports the management of machine learning models. 
   3. Liya Yuan: Since several meetups and outreach activities are yet to come, detailed document need to be prepared in advance.

12/17

1. Roll Call

1. 	Company	Attendee
   TSC members	ZTE	Tao Liu, Shiying Jin, Liya Yuan
   	China Mobile	Qi Sun
   	China Unicom	Tengfei Liu
   Other attendees

2. Adlik TSC chair election

1. TSC voted and approved Liya Yuan as Adlik TSC chair.

3. Status of open action items

1. Further discussion on ingration of Adlik in MEC. ----- Discussion materials will be provided and discussed via emails first.

2. Documentation for detailed information about how to use the compiler. ----- Will be included as a new feature.

4. Work report and plans

1. Roadmap is updated on Github.

2. FPGA compiler and runtime will be developed first.

5. Open Discussion

1. Call for more developers to contribute to Adlik, e.g. to enable support for more op.

12/03

1. Self introdution of TSC members and developers

1. 	Company	Attendee
   TSC members	ZTE	Tao Liu, Shiying Jin, Liya Yuan
   	China Mobile	Qi Sun
   	China Unicom	Tengfei Liu, Ao Li
   Other attendees		Vishnu Ram

2. Adlik code walkthrough

1. Introduction to the two sub projects, model compiler and model serving(Tao Liu and Shiying Jin).

1.  Q: How to give input to the model compiler? A: It's shown in the simpleMNIST demo.

2.  Q: I have to always use the name mnist.h5 and manually input the layers in this compile_model.py? A: Further documentation will show how to get these parameters and pass them to compiler.

3.  Adlik work plans

1.  New features in progress: support for ARM CPU and FPGA; support for quantilization(Tao Liu).

4. Open Discussion

1. Adlik intergration with MEC(Ao Li)

1. The interface requirements need further discussion.

2. Requirements for APIs can be written collaboratively with ITU-T FG ML5G.

5. Actions

1. Further discussion on ingration of Adlik in MEC.

2. Documentation for detailed information about how to use the compiler.