Page tree

Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

Event planning and general information template below - To be updated by Event/Meeting Lead

Table of Contents

Event/Meeting Details

LF AI & Data Participation Overview: 

LF AI & Data Foundation is pleased to welcome you at the LF AI & Data Day* – ONNX Community Virtual ONNX Runtime Training Update.mp4 Meetup – March 2021. This event will be hosted online via Zoom video conferencing on Wednesday, March 24. The event will feature the LF AI & Data hosted Graduated Project, ONNX.

The event will cover ONNX Community updates, partner/end-user stories, and SIG/WG updates. Check back for a full agenda soon.

If you are using ONNX in your services and applications, building software or hardware that supports ONNX, or contributing to ONNX, you should attend! This is a great opportunity to meet with and hear from people working with ONNX across many companies.

Note: In order to ensure the safety of our event participants and staff due to the Novel Coronavirus situation (COVID-19) the ONNX Steering Committee decided to make this a virtual-only event via Zoom.

*LF AI & Data Day is a regional, one-day event hosted and organized by local members with support from LF AI & Data and its Projects. Learn more about the LF AI & Data Foundation here.

Event Location & Date(s): 

Zoom meeting on Wednesday, March 24, 2021

Event Time:

5:00 PM - 8:00 PM - PDT (Pacific Daylight Time)

Event Website: https://events.linuxfoundation.org/lf-ai-data-day-onnx-community-virtual-meetup-spring/ 

Event Registration: https://www.cvent.com/d/37qh6p/4W?ct=50221cf5-5496-4c34-9ec0-3b52b1bf1204&_ga=2.191814634.1770376097.1607959940-1411187557.1597333659

Event Agenda/Schedule: https://events.linuxfoundation.org/lf-ai-data-day-onnx-community-virtual-meetup-spring/program/schedule/

Please submit any questions about LF AI & Data participation at this event to: outreach-committee@lists.lfaidata.foundation

Event Host Lead Name & Contact Details: Ti Zhou (Baidu) <zhouti@baidu.com>

Planning To-Do List

Tracking for to-do items, due dates, owners, and notes

To Do

Due Date

Completed

Owner

Notes

Post presentations and recordingsMarch 25
Event Lead
Communicate availability of the event's recordings and presentationsMarch 31
LF AI & Data-LF AI social scheduled for TBD.

Schedule & Presentations

Time

Duration

Topic / Speaker

Deck / Recording Links

Wed March 24
5pm PT USA

Thur March 25
8am China

25mEvent Kickoff

Q&A - #onnx-general slack channel

For those speakers who would like to - they are invited to make both English (required) and Chinese (optional) versions of their deck and pre-recording

post-event everyone pointed to this wiki page for full recording


5Welcome and opening
Speaker: Ti Zhou (Baidu)

deck / recording


20

ONNX Progress Update
Speaker: Sheng Zha, Joohoon Lee
Contacts: ONNX SC (Prasanth, Harry, Jim, Joohoon, Sheng)

deck / recording (Sheng) / recording (Joohoon)

5:25pm PT USA

8:25am China

100mCommunity Presentations

deck / prerecording (English)

optional: deck/record(Chinese)


10m

popART: Support ONNX on IPU
Speaker: Han Zhao (GraphCore-China)


10m

Spring Project:Multi Backend Neural Network Auto Quantization and Deployment over ONNX
Speaker: Fengwei Yu (SenseTime-China)


10m

ONNX Runtime for Mobile Scenarios: From model to on-device inferencing
Speaker: Tom Wildenhain (Microsoft-USA) and Scott McKay (Microsoft-Australia)


10m

ONNX on PaddlePaddle 2.0: Broader Deployment and Richer Ecosystem
Speaker: Wranky Wang (Baidu-China)


10m

ONNX on microcontrollers
Speaker: Rohit Sharma (AITechSystems-USA_CA)


10m

Monitoring and Explaining ONNX Models in Production
Speaker: Krishna Gade (FiddlerAI-USA_CA)

deckrecording

10mONNX client for Acumos
Speaker: Philippe Dooze (Orange-France) (picture)

10m

Deploy ONNX model seamlessly across the cloud, edge, and mobile devices using MindSpore
Speaker: Leon Wang (Huawei-China)


10m

ONNX Runtime Training
Speaker: Peng Wang (Microsoft China)


10m

Quantization support for ONNX using LPOT (Low precision optimization tool)

Speakers: Haihao Shen (Intel - China) and Saurabh Tangri (Intel)
Contact: Rajeev Nalawadi (Intel-China)

deck / recording

7:05pm PT USA

10:05am China

10mBreakQ&A - #onnx-general slack channel

7:15pm PT USA

10:15am China

45mSIG's and WG's Updates and Discussions

deck / prerecording (English)

optional:deck/record (Chinese)


10Architecture/Infrastructure SIG Update
Chair: Ashwini Khade (Microsoft) and Jacky Chen (Microsoft)

10Operators SIG Update
Co-Chairs: Michał Karzyński (Intel) and G. Ramalingam (Microsoft)

10

Converters SIG Update
Co-Chairs: Guenther Schmuelling (Microsoft), Kevin Chen (Nvidia), Chin Huang (IBM)


10

Model Zoo/Tutorials SIG Update
Co-Chair: Wenbing Li (Microsoft)(picture)


5Q&A / Open DiscussionsQ&A - #onnx-general slack channel

8:00pm PT USA

11:00am China

Total 180

Minutes

EndFinal Recording Full Event (Posted after online event)



Q&A

SpeakerQuestionAnswer
Welcome

Ti ZhouHow many models in the PaddlePaddle model zoo?  How many in ONNX format?
Sheng ZhaPlease remind where to find information on upcoming ONNX SC elections?
Joohoon LeeWhat is the best way to get started learning about the ONNX roadmap, and getting involved?
Community Talks

Han Zhao (GraphCore-China)Are an performance benchmarks available on ML Perf/ML Commons?
Fengwei Yu (SenseTime-China)

Tom Wildenhain (Microsoft-USA) and Scott McKay (Microsoft-Australia)Are any ONNX models being run a lot on smartphones today that you know of?
Wranky Wang (Baidu-China)Are there plans to put PaddlePaddle into open governance in LF AI & Data?
Rohit Sharma (AITechSystems-USA_CA)How many of the models in ONNX model zoo have you compiled onto tinyml devices using DeepC?  of the 36, have 2-3 been compiled? on what device?

Over 50 ONNX models have been compiled. We do not keep the record of the source of those ONNX models, so we can't comment on how many of them come from ONNX model zoo. Some of them are available in cAInvas gallery

deepSea supports most RISC and CISC devices. Cortex M4 based boards like Arduino Nano is one of the popular ones in the gallery

Krishna Gade (FiddlerAI-USA_CA)Why is model versioning a big problem?  Of the models in the ONNX model zoo, which are black box and which are explainable, from a Fiddler perspective?
Philippe Dooze (Orange-France) (picture)Who are the biggest users of Acumos today?
Leon Wang (Huawei-China)Are any unique models in the ONNX model zoo currently being developed with Mindspore?
Peng Wang (Microsoft China)What are the challenges of building an training graph from existing ONNX models in the ONNX model zoo?
Speakers: Haihao Shen (Intel - China) and Saurabh Tangri (Intel)
Contact: Rajeev Nalawadi (Intel-China)
Is LPOT open source?  https://github.com/intel/lpot  Are there plans to move it to LF AI & Data?
SIG Talks

Ashwini Khade (Microsoft) and Jacky Chen (Microsoft)

Michał Karzyński (Intel) and G. Ramalingam (Microsoft)

Guenther Schmuelling (Microsoft), Kevin Chen (Nvidia), Chin Huang (IBM)

Model Zoo/Tutorials SIG Update
Co-Chair: Wenbing Li (Microsoft)(picture)
When do you think the number of models in the ONNX model zoo will be much greater than the number of ONNX operators?

Additional Community Submissions not in main program  - not covered online, but thank-you to submitters who can upload deck and prerecording for community to review at their leisure

Session

Deck / Prerecording (Links)

My experience implementing ONNX import for GAP processors
Contact: Martin Croome (Greenwaves Tech)

How we are making it insanely easy to deploy ml/ai models from jetson nano to Azure with ONNX
Contact: Mahesh Yadav (Microsoft)
deck / prerecording
Visualizing ONNX models’ internal data: Key things to look for?
Contact: Mina Amiri (Zetane)

Deploying 3rd Party Models in PaddlePaddle via X2Paddle Converter
Contact: JiaJun Jiang (Baidu)

deck / prerecording

TBD

Contact: Emad Barsoum (Cerebras) 

deck / prerecording