[jira] [Created] (MXNET-688) Fix quantization divide by zero errors
Ray Zhang created MXNET-688: --- Summary: Fix quantization divide by zero errors Key: MXNET-688 URL: https://issues.apache.org/jira/browse/MXNET-688 Project: Apache MXNet Issue Type: Bug Reporter: Ray Zhang The current quantization strategy for `calib_mode='entropy'` is to calculate the KL divergence for different thresholds and choose the best threshold. This assumes that the random variable is nonzero for all reals and is a continuous random variable. Because we are discretizing the distribution, we smooth the distribution over the range `[-threshold, threshold]`. What we are not considering is that the entire sampled distribution may be not in the range `[-threshold, threshold]` and thus we end up with all zeros in the sampled candidate `p` distribution inside of `_get_optimal_threshold`. I have added a check that the distribution(possibly unnormalized) is proper before attempting to smooth or else we'll run into a divide by 0 error. In most cases, activation functions and layers for classification type problems output numbers symmetric around 0. This is not the case for a regressor's last layer, and there are various other examples where the activation distribution is not around 0, and this was a major blockage for airbnb's adoption into mxnet's quantization capabilities. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@mxnet.apache.org For additional commands, e-mail: issues-h...@mxnet.apache.org
[jira] [Updated] (MXNET-139) Tutorial for FP16
[ https://issues.apache.org/jira/browse/MXNET-139?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Rahul Huilgol updated MXNET-139: Status: Done (was: In Review) > Tutorial for FP16 > - > > Key: MXNET-139 > URL: https://issues.apache.org/jira/browse/MXNET-139 > Project: Apache MXNet > Issue Type: Improvement >Reporter: Rahul Huilgol >Assignee: Rahul Huilgol >Priority: Major > Time Spent: 6h > Remaining Estimate: 0h > -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@mxnet.apache.org For additional commands, e-mail: issues-h...@mxnet.apache.org
[jira] [Updated] (MXNET-484) Add C++ unit tests for lrn operator
[ https://issues.apache.org/jira/browse/MXNET-484?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Alexander updated MXNET-484: Status: In Review (was: In Progress) > Add C++ unit tests for lrn operator > --- > > Key: MXNET-484 > URL: https://issues.apache.org/jira/browse/MXNET-484 > Project: Apache MXNet > Issue Type: Sub-task >Reporter: Alexander >Assignee: Alexander >Priority: Minor > -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@mxnet.apache.org For additional commands, e-mail: issues-h...@mxnet.apache.org
[jira] [Commented] (MXNET-661) GPU throws out of index error?
[ https://issues.apache.org/jira/browse/MXNET-661?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16549969#comment-16549969 ] Vandana Kannan commented on MXNET-661: -- Submitter resolved issue by setting appropriate context, but suggested that there should be a setting for global context or the ability to switch contexts automatically. Asked the submitter if issue can be closed. The feature request to switch context automatically was initiated in a separate thread ([https://github.com/apache/incubator-mxnet/issues/6333).|https://github.com/apache/incubator-mxnet/issues/6333.] > GPU throws out of index error? > -- > > Key: MXNET-661 > URL: https://issues.apache.org/jira/browse/MXNET-661 > Project: Apache MXNet > Issue Type: Bug >Reporter: ROSHANI NAGMOTE >Assignee: Vandana Kannan >Priority: Major > > https://github.com/apache/incubator-mxnet/issues/8794 -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@mxnet.apache.org For additional commands, e-mail: issues-h...@mxnet.apache.org
[jira] [Commented] (MXNET-579) Gluon raises error if the user does not call nd.waitall()
[ https://issues.apache.org/jira/browse/MXNET-579?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16549826#comment-16549826 ] Lin Yuan commented on MXNET-579: No longer reproducible > Gluon raises error if the user does not call nd.waitall() > - > > Key: MXNET-579 > URL: https://issues.apache.org/jira/browse/MXNET-579 > Project: Apache MXNet > Issue Type: Bug >Reporter: Lin Yuan >Assignee: Lin Yuan >Priority: Major > -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@mxnet.apache.org For additional commands, e-mail: issues-h...@mxnet.apache.org
[jira] [Updated] (MXNET-579) Gluon raises error if the user does not call nd.waitall()
[ https://issues.apache.org/jira/browse/MXNET-579?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Lin Yuan updated MXNET-579: --- Status: In Progress (was: To Do) > Gluon raises error if the user does not call nd.waitall() > - > > Key: MXNET-579 > URL: https://issues.apache.org/jira/browse/MXNET-579 > Project: Apache MXNet > Issue Type: Bug >Reporter: Lin Yuan >Assignee: Lin Yuan >Priority: Major > -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@mxnet.apache.org For additional commands, e-mail: issues-h...@mxnet.apache.org
[jira] [Commented] (MXNET-577) Crash while running gluon image-classification.py example with float16
[ https://issues.apache.org/jira/browse/MXNET-577?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16549811#comment-16549811 ] Lin Yuan commented on MXNET-577: The issue has been fixed in the latest release. Verified on p2.8xlarge instance: python example/gluon/image_classification.py --dataset dummy --gpus 0 --epochs 10 --mode imperative --model resnet50_v2 --batch-size 128 --log-interval 1 --dtype float16 > Crash while running gluon image-classification.py example with float16 > -- > > Key: MXNET-577 > URL: https://issues.apache.org/jira/browse/MXNET-577 > Project: Apache MXNet > Issue Type: Bug >Reporter: Lin Yuan >Assignee: Lin Yuan >Priority: Major > -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@mxnet.apache.org For additional commands, e-mail: issues-h...@mxnet.apache.org
[jira] [Updated] (MXNET-577) Crash while running gluon image-classification.py example with float16
[ https://issues.apache.org/jira/browse/MXNET-577?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Lin Yuan updated MXNET-577: --- Status: In Progress (was: To Do) > Crash while running gluon image-classification.py example with float16 > -- > > Key: MXNET-577 > URL: https://issues.apache.org/jira/browse/MXNET-577 > Project: Apache MXNet > Issue Type: Bug >Reporter: Lin Yuan >Assignee: Lin Yuan >Priority: Major > -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@mxnet.apache.org For additional commands, e-mail: issues-h...@mxnet.apache.org
[jira] [Updated] (MXNET-560) Adding a temperature parameter in Softmax and SoftmaxOutput operator
[ https://issues.apache.org/jira/browse/MXNET-560?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Lin Yuan updated MXNET-560: --- Resolution: Done Status: Done (was: To Do) PR is merged: https://github.com/apache/incubator-mxnet/pull/11466 Ready for testing. > Adding a temperature parameter in Softmax and SoftmaxOutput operator > > > Key: MXNET-560 > URL: https://issues.apache.org/jira/browse/MXNET-560 > Project: Apache MXNet > Issue Type: New Feature >Reporter: Lin Yuan >Assignee: Lin Yuan >Priority: Major > Time Spent: 12h 40m > Remaining Estimate: 0h > > MXNet does not appear to have a native temperature parameter in its softmax > functions. I would like this to be added, as it has many useful applications > when learning a categorical probability distribution, especially in a > reinforcement learning setting. Should default to 1 to reproduce current > behavior. > [https://en.wikipedia.org/wiki/Softmax_function#Reinforcement_learning] -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@mxnet.apache.org For additional commands, e-mail: issues-h...@mxnet.apache.org
[jira] [Assigned] (MXNET-560) Adding a temperature parameter in Softmax and SoftmaxOutput operator
[ https://issues.apache.org/jira/browse/MXNET-560?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Lin Yuan reassigned MXNET-560: -- Assignee: Lin Yuan > Adding a temperature parameter in Softmax and SoftmaxOutput operator > > > Key: MXNET-560 > URL: https://issues.apache.org/jira/browse/MXNET-560 > Project: Apache MXNet > Issue Type: New Feature >Reporter: Lin Yuan >Assignee: Lin Yuan >Priority: Major > Time Spent: 12h 40m > Remaining Estimate: 0h > > MXNet does not appear to have a native temperature parameter in its softmax > functions. I would like this to be added, as it has many useful applications > when learning a categorical probability distribution, especially in a > reinforcement learning setting. Should default to 1 to reproduce current > behavior. > [https://en.wikipedia.org/wiki/Softmax_function#Reinforcement_learning] -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@mxnet.apache.org For additional commands, e-mail: issues-h...@mxnet.apache.org
[jira] [Commented] (MXNET-656) Bug in mxnet.contrib.text.utils.count_tokens_from_str
[ https://issues.apache.org/jira/browse/MXNET-656?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16549441#comment-16549441 ] Vandana Kannan commented on MXNET-656: -- API documentation updated > Bug in mxnet.contrib.text.utils.count_tokens_from_str > -- > > Key: MXNET-656 > URL: https://issues.apache.org/jira/browse/MXNET-656 > Project: Apache MXNet > Issue Type: Bug >Reporter: ROSHANI NAGMOTE >Assignee: Vandana Kannan >Priority: Major > Time Spent: 1h > Remaining Estimate: 0h > > https://github.com/apache/incubator-mxnet/issues/11526 -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@mxnet.apache.org For additional commands, e-mail: issues-h...@mxnet.apache.org
[jira] [Commented] (MXNET-667) AssertionError: Data must be list of NDArrays
[ https://issues.apache.org/jira/browse/MXNET-667?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16549436#comment-16549436 ] Vandana Kannan commented on MXNET-667: -- Issue does not occur on latest MXNet code > AssertionError: Data must be list of NDArrays > - > > Key: MXNET-667 > URL: https://issues.apache.org/jira/browse/MXNET-667 > Project: Apache MXNet > Issue Type: Bug >Reporter: ROSHANI NAGMOTE >Assignee: Vandana Kannan >Priority: Major > > https://github.com/apache/incubator-mxnet/issues/7784 -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@mxnet.apache.org For additional commands, e-mail: issues-h...@mxnet.apache.org
[jira] [Commented] (MXNET-665) use 224x224 size to train mobilenet will encounter memory problem
[ https://issues.apache.org/jira/browse/MXNET-665?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16549435#comment-16549435 ] Vandana Kannan commented on MXNET-665: -- The size mismatch error does not occur with the latest MXNet code. > use 224x224 size to train mobilenet will encounter memory problem > - > > Key: MXNET-665 > URL: https://issues.apache.org/jira/browse/MXNET-665 > Project: Apache MXNet > Issue Type: Bug >Reporter: ROSHANI NAGMOTE >Assignee: Vandana Kannan >Priority: Major > > https://github.com/apache/incubator-mxnet/issues/8391 -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@mxnet.apache.org For additional commands, e-mail: issues-h...@mxnet.apache.org
