Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
    Alex Merritt
    @EmergentOrder
    @vinitra I'm on the community meetup, it was mentioned earlier that the models repo has added GPT-2 & NCF (Neural Collaborative Filtering) since the last release. I found GPT-2, nice to have that. I don't see anything but the stub for NCF though, is there WIP there or was that a mistake?
    Sree Vaddi
    @sreev
    hi
    @prasanthpul adding checksum , adding additional layer of security doesn't hurt. but is it valid , though ?
    a better checksum can be, imho:
    validity of the data used in making the model.
    validity of model itself.
    Sree Vaddi
    @sreev
    eg. download a picture of acquarium , use this against this model, it should identify 90% of the fish ?
    or on that lines...
    Vinitra Swamy
    @vinitra
    @EmergentOrder Hi Alex. I highlighted the Fast Neural Style Transfer and Super Resolution models in addition to GPT-2 during the ONNX workshop. I'm not aware of work on NCF currently, although we always welcome contributions! Which talk mentioned it?
    Vinitra Swamy
    @vinitra

    @/all The next SIG meeting has been moved to this Monday, 4/13 at 2 - 2:30 PM PST. Hope to see you all there.

    Topic: Model Zoo + Tutorials Meeting #3
    Time: Apr 13, 2020 02:00 PM Pacific Time (US and Canada)

    https://zoom.us/j/341887317
    Meeting ID: 341 887 317

    Alex Merritt
    @EmergentOrder
    @vinitra It was in the update at the beginning, I guess they must have confused NCF and Fast Neural Style Transfer. no worries, just wanted to clarify that
    Vinitra Swamy
    @vinitra
    Reminder -- meeting today at 2! We're tracking several issues here, would love community input on these topics (opset versioning, model zoo CI, github analytics): https://github.com/onnx/sigs/issues
    1 reply
    Zhipeng Huang
    @hannibalhuang
    quick question, any specific reason the modelzoo is licensed under MIT instead of Apache or BSD ?
    Vinitra Swamy
    @vinitra
    @hannibalhuang Some models are contributed by users / companies who prefer a specific license -- you will see a variety of licenses throughout the zoo. Do you have a preference? We can look into standardization.
    Prasanth Pulavarthi
    @prasanthpul
    everything in ONNX is under MIT license for usage, with a CLA for contributions. with the move to LF AI, we will be moving to Apache 2.0 license with DCO for contributions (announced at recent workshop).
    daquexian
    @daquexian
    Hi, I think the resnetv1 onnx model in model zoo https://github.com/onnx/models/tree/master/vision/classification/resnet is wrong. The spatial attr of batchnorm is incorrectly set to 0 and as a result an error is raised when running this model in onnxruntime
    Vinitra Swamy
    @vinitra
    Hi @daquexian -- yes, that's a known issue! Thanks for reminding us. I fixed this on the MXNet exporter side: apache/incubator-mxnet#17711 a few weeks ago, but forgot to update the models. Will make a PR shortly.
    Vinitra Swamy
    @vinitra
    Zhipeng Huang
    @hannibalhuang
    @prasanthpul does it mean Apache also applied to modelzoo ? I've been doing some research on modelzoo licensing but it seems like there is no common practice
    Michał Karzyński
    @postrational
    There seem to be some malformed tar files in the onnx/models LFS archives:
    $ wget https://media.githubusercontent.com/media/onnx/models/master/vision/classification/resnet/model/resnet101-v1-7.tar.gz
    --2020-05-21 15:48:44--  https://media.githubusercontent.com/media/onnx/models/master/vision/classification/resnet/model/resnet101-v1-7.tar.gz
    Proxy request sent, awaiting response... 200 OK
    Length: 167595867 (160M) [application/octet-stream]
    Saving to: ‘resnet101-v1-7.tar.gz’
    
    resnet101-v1-7.tar.gz  100%[===========================================>] 159.83M  5.29MB/s    in 24s
    
    $ tar -xzvf resnet101-v1-7.tar
    x resnet101v1/
    x resnet101v1/resnet101-v1.onnx
    x resnet101v1/test_data_set_0/
    x resnet101v1/test_data_set_0/._input_0.pb
    x resnet101v1/test_data_set_0/._output_0.pb
    x resnet101v1/test_data_set_0/input_0.pb
    x resnet101v1/test_data_set_0/output_0.pb
    tar: Ignoring malformed pax extended attributeIgnoring malformed pax extended attributeIgnoring malformed pax extended attribute
    x resnet101v1/test_data_set_1/
    x resnet101v1/test_data_set_1/._input_0.pb
    x resnet101v1/test_data_set_1/._output_0.pb
    x resnet101v1/test_data_set_1/input_0.pb
    x resnet101v1/test_data_set_1/output_0.pb
    tar: Ignoring malformed pax extended attributeIgnoring malformed pax extended attributeIgnoring malformed pax extended attribute
    x resnet101v1/test_data_set_2/
    x resnet101v1/test_data_set_2/._input_0.pb
    x resnet101v1/test_data_set_2/._output_0.pb
    x resnet101v1/test_data_set_2/input_0.pb
    x resnet101v1/test_data_set_2/output_0.pb
    tar: Damaged tar archive
    tar: Error exit delayed from previous errors.
    Michał Karzyński
    @postrational
    It seems that all of these models have malformed tar files (most ResNet models):
    onnx/models/master/vision/classification/resnet/model/resnet101-v1-7.tar.gz
    onnx/models/master/vision/classification/resnet/model/resnet101-v2-7.tar.gz
    onnx/models/master/vision/classification/resnet/model/resnet152-v1-7.tar.gz
    onnx/models/master/vision/classification/resnet/model/resnet152-v2-7.tar.gz
    onnx/models/master/vision/classification/resnet/model/resnet18-v2-7.tar.gz
    onnx/models/master/vision/classification/resnet/model/resnet34-v1-7.tar.gz
    onnx/models/master/vision/classification/resnet/model/resnet34-v2-7.tar.gz
    onnx/models/master/vision/classification/resnet/model/resnet50-v1-7.tar.gz
    onnx/models/master/vision/classification/resnet/model/resnet50-v2-7.tar.gz
    Michał Karzyński
    @postrational
    I created an issue to track this: onnx/models#309
    @vinitra Could you take a look?
    Vinitra Swamy
    @vinitra
    Addressed in recent PR onnx/models#312. Thanks for the bug report @postrational.

    Next meeting will be mid-June! Some quick updates --

    Reviews and community contributions are encouraged and welcome. :)

    Vinitra Swamy
    @vinitra
    Next SIG meeting - Monday, June 15, 1 - 1:30 PM PST. Looking forward to seeing you there! https://zoom.us/j/96099632051
    2 replies
    kumariKambham
    @kumariKambham
    Hi, I am new to this . Can you help me ?
    8 replies
    kumariKambham
    @kumariKambham
    Hi, Is there any possibilities to learn ML with python
    1 reply
    Vinitra Swamy
    @vinitra
    Lots of exciting new updates in the Model Zoo! Next meeting will be Friday, August 14th from 11 - 11:30 AM PST. See you there! https://zoom.us/j/97523672358
    Vinitra Swamy
    @vinitra
    Next Model Zoo SIG meeting will be Friday, Sept 25th at 10:30 AM - 11 AM PST! A few updates -- I'll be leaving for a PhD program in Switzerland, so I'm handing off the SIG lead role to Wenbing Li, the top contributor to the ONNX converters effort. We're also looking for a co-lead for this SIG. Contact the ONNX steering committee if interested.
    3 replies
    Harry Kim
    @harryskim
    We are migrating Gitter conversations to Slack. Please sign up for LF AI slack using https://slack.lfai.foundation/ and join "onnx-modelzoo" channel for the model zoo SIG (https://lfaifoundation.slack.com/archives/C018RE2BRBL).
    Prasanth Pulavarthi
    @prasanthpul
    posting in all channels:
    ONNX will be switching to the Apache 2.0 license by the end of this month. Instead of signing a CLA, contributors need to provide a DCO (developer certificate of origin). This is done by including a sign-off-by line in commit messages. Using the “-s” flag for “git commit” will automatically append this line. For example, running “git commit -s -m ‘commit info.’” it will produce a commit that has the message “commit info. Signed-off-by: First Last email@company.com”. The DCO bot will ensure commits are signed with an email address that matches the commit author before they are eligible to be merged.