This is an automated email from the ASF dual-hosted git repository. aaronmarkham pushed a commit to branch asf-site in repository https://gitbox.apache.org/repos/asf/incubator-mxnet-site.git
The following commit(s) were added to refs/heads/asf-site by this push: new ecf65d0 Publish triggered by CI ecf65d0 is described below commit ecf65d0bca438dc3eae10fafa5011ff53032a78b Author: mxnet-ci <mxnet-ci> AuthorDate: Thu Jul 2 18:41:10 2020 +0000 Publish triggered by CI --- api/python/docs/tutorials/packages/gluon/image/info_gan.html | 4 ++-- .../packages/gluon/training/normalization/index.html | 12 ++++++------ api/python/docs/tutorials/performance/backend/profiler.html | 8 ++++---- date.txt | 1 - feed.xml | 2 +- 5 files changed, 13 insertions(+), 14 deletions(-) diff --git a/api/python/docs/tutorials/packages/gluon/image/info_gan.html b/api/python/docs/tutorials/packages/gluon/image/info_gan.html index f8fe089..fd806fb 100644 --- a/api/python/docs/tutorials/packages/gluon/image/info_gan.html +++ b/api/python/docs/tutorials/packages/gluon/image/info_gan.html @@ -908,9 +908,9 @@ notebook uses the DCGAN example from the <a class="reference external" href="htt </pre></div> </div> <p>There are 2 differences between InfoGAN and DCGAN: the extra latent code and the Q network to estimate the code. The latent code is part of the Generator input and it contains mutliple variables (continuous, categorical) that can represent different distributions. In order to make sure that the Generator uses the latent code, mutual information is introduced into the GAN loss term. Mutual information measures how much X is known given Y or vice versa. It is defined as:</p> -<p><img alt="gif" src="https://raw.githubusercontent.com/dmlc/web-data/master/mxnet/doc/tutorials/info_gan/loss.gif" /></p> +<p><img alt="gif" src="https://raw.githubusercontent.com/dmlc/web-data/master/mxnet/doc/tutorials/info_gan/entropy.gif" /></p> <p>The InfoGAN loss is:</p> -<p><img alt="gif" src="https://raw.githubusercontent.com/dmlc/web-data/master/mxnet/doc/tutorials/info_gan/loss.gif" /></p> +<p><img alt="image1" src="https://raw.githubusercontent.com/dmlc/web-data/master/mxnet/doc/tutorials/info_gan/loss.gif" /></p> <p>where <code class="docutils literal notranslate"><span class="pre">V(D,G)</span></code> is the GAN loss and the mutual information <code class="docutils literal notranslate"><span class="pre">I(c,</span> <span class="pre">G(z,</span> <span class="pre">c))</span></code> goes in as regularization. The goal is to reach high mutual information, in order to learn meaningful codes for the data.</p> <p>Define the loss functions. <code class="docutils literal notranslate"><span class="pre">SoftmaxCrossEntropyLoss</span></code> for the categorical code, <code class="docutils literal notranslate"><span class="pre">L2Loss</span></code> for the continious code and <code class="docutils literal notranslate"><span class="pre">SigmoidBinaryCrossEntropyLoss</span></code> for the normal GAN loss.</p> <div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">loss1</span> <span class="o">=</span> <span class="n">gluon</span><span class="o">.</span><span class="n">loss</span><span class="o">.</span><span class="n">SigmoidBinaryCrossEntropyLoss</span><span class="p">()</span> diff --git a/api/python/docs/tutorials/packages/gluon/training/normalization/index.html b/api/python/docs/tutorials/packages/gluon/training/normalization/index.html index 9fe193f..33532e1 100644 --- a/api/python/docs/tutorials/packages/gluon/training/normalization/index.html +++ b/api/python/docs/tutorials/packages/gluon/training/normalization/index.html @@ -772,8 +772,8 @@ shifting certain values towards a distribution with a mean of 0 (i.e. zero-cent </tr> </thead> <tbody> -<tr class="row-even"><td><p><img alt="image0" src="tutorials/packages/gluon/training/normalization/imgs/NCHW_IN.png" /></p></td> -<td><p><img alt="image1" src="tutorials/packages/gluon/training/normalization/imgs/NTC_IN.png" /></p></td> +<tr class="row-even"><td><p><img alt="image1" src="tutorials/packages/gluon/training/normalization/imgs/NCHW_IN.png" /></p></td> +<td><p><img alt="image2" src="tutorials/packages/gluon/training/normalization/imgs/NTC_IN.png" /></p></td> </tr> <tr class="row-odd"><td><p>(e.g. batch of images) using the default of <code class="docutils literal notranslate"><span class="pre">axis=1</span></code></p></td> <td><p>(e.g. batch of sequences) overriding the default with <code class="docutils literal notranslate"><span class="pre">axis=2</span></code> (or <code class="docutils literal notranslate"><span class="pre">axis=-1</span></code>)</p></td> @@ -880,8 +880,8 @@ to adjust to shifts in the input distribution. Using the same batch another 100 </tr> </thead> <tbody> -<tr class="row-even"><td><p><img alt="image0" src="tutorials/packages/gluon/training/normalization/imgs/NCHW_IN.png" /></p></td> -<td><p><img alt="image1" src="tutorials/packages/gluon/training/normalization/imgs/NTC_IN.png" /></p></td> +<tr class="row-even"><td><p><img alt="image1" src="tutorials/packages/gluon/training/normalization/imgs/NCHW_IN.png" /></p></td> +<td><p><img alt="image2" src="tutorials/packages/gluon/training/normalization/imgs/NTC_IN.png" /></p></td> </tr> <tr class="row-odd"><td><p>(e.g. batch of images) overriding the default with <code class="docutils literal notranslate"><span class="pre">axis=1</span></code></p></td> <td><p>(e.g. batch of sequences) using the default of <code class="docutils literal notranslate"><span class="pre">axis=-1</span></code></p></td> @@ -922,8 +922,8 @@ to adjust to shifts in the input distribution. Using the same batch another 100 </tr> </thead> <tbody> -<tr class="row-even"><td><p><img alt="image0" src="tutorials/packages/gluon/training/normalization/imgs/NCHW_IN.png" /></p></td> -<td><p><img alt="image1" src="tutorials/packages/gluon/training/normalization/imgs/NTC_IN.png" /></p></td> +<tr class="row-even"><td><p><img alt="image1" src="tutorials/packages/gluon/training/normalization/imgs/NCHW_IN.png" /></p></td> +<td><p><img alt="image2" src="tutorials/packages/gluon/training/normalization/imgs/NTC_IN.png" /></p></td> </tr> <tr class="row-odd"><td><p>(e.g. batch of images) using the default <code class="docutils literal notranslate"><span class="pre">axis=1</span></code></p></td> <td><p>(e.g. batch of sequences) overiding the default with <code class="docutils literal notranslate"><span class="pre">axis=2</span></code> (or <code class="docutils literal notranslate"><span class="pre">axis=-1</span></code> equivalently)</p></td> diff --git a/api/python/docs/tutorials/performance/backend/profiler.html b/api/python/docs/tutorials/performance/backend/profiler.html index d76a077..bf772ab 100644 --- a/api/python/docs/tutorials/performance/backend/profiler.html +++ b/api/python/docs/tutorials/performance/backend/profiler.html @@ -886,7 +886,7 @@ profiling jointly with vendor tools is recommended.</p> <p><code class="docutils literal notranslate"><span class="pre">dump()</span></code> creates a <code class="docutils literal notranslate"><span class="pre">json</span></code> file which can be viewed using a trace consumer like <code class="docutils literal notranslate"><span class="pre">chrome://tracing</span></code> in the Chrome browser. Here is a snapshot that shows the output of the profiling we did above. Note that setting the <code class="docutils literal notranslate"><span class= [...] <p><img alt="Tracing Screenshot" src="https://raw.githubusercontent.com/dmlc/web-data/master/mxnet/tutorials/python/profiler/profiler_output_chrome.png" /></p> <p>Let’s zoom in to check the time taken by operators</p> -<p><img alt="Operator profiling" src="https://raw.githubusercontent.com/dmlc/web-data/master/mxnet/tutorials/python/profiler/profiler_winograd.png" /></p> +<p><img alt="Operator profiling" src="https://raw.githubusercontent.com/dmlc/web-data/master/mxnet/tutorials/python/profiler/profiler_nvprof.png" /></p> <p>The above picture visualizes the sequence in which the operators were executed and the time taken by each operator.</p> </div> </div> @@ -993,11 +993,11 @@ scripts running MXNet. And you can use these in conjunction with the MXNet Profi <p><code class="docutils literal notranslate"><span class="pre">==11588==</span> <span class="pre">Generated</span> <span class="pre">result</span> <span class="pre">file:</span> <span class="pre">/home/user/Development/incubator-mxnet/ci/my_profile.nvvp</span></code></p> <p>We specified an output file called <code class="docutils literal notranslate"><span class="pre">my_profile.nvvp</span></code> and this will be annotated with NVTX ranges (for MXNet operations) that will be displayed alongside the standard NVProf timeline. This can be very useful when you’re trying to find patterns between operators run by MXNet, and their associated CUDA kernel calls.</p> <p>You can open this file in Visual Profiler to visualize the results.</p> -<p><img alt="Operator profiling" src="https://raw.githubusercontent.com/dmlc/web-data/master/mxnet/tutorials/python/profiler/profiler_winograd.png" /></p> +<p><img alt="Operator profiling" src="https://raw.githubusercontent.com/dmlc/web-data/master/mxnet/tutorials/python/profiler/profiler_nvprof.png" /></p> <p>At the top of the plot we have CPU tasks such as driver operations, memory copy calls, MXNet engine operator invocations, and imperative MXNet API calls. Below we see the kernels active on the GPU during the same time period.</p> -<p><img alt="Operator profiling" src="https://raw.githubusercontent.com/dmlc/web-data/master/mxnet/tutorials/python/profiler/profiler_winograd.png" /></p> +<p><img alt="image1" src="https://raw.githubusercontent.com/dmlc/web-data/master/mxnet/tutorials/python/profiler/profiler_nvprof_zoomed.png" /></p> <p>Zooming in on a backwards convolution operator we can see that it is in fact made up of a number of different GPU kernel calls, including a cuDNN winograd convolution call, and a fast-fourier transform call.</p> -<p><img alt="Operator profiling" src="https://raw.githubusercontent.com/dmlc/web-data/master/mxnet/tutorials/python/profiler/profiler_winograd.png" /></p> +<p><img alt="image2" src="https://raw.githubusercontent.com/dmlc/web-data/master/mxnet/tutorials/python/profiler/profiler_winograd.png" /></p> <p>Selecting any of these kernel calls (the winograd convolution call shown here) will get you some interesting GPU performance information such as occupancy rates (vs theoretical), shared memory usage and execution duration.</p> <p>Nsight Compute is available in CUDA 10 toolkit, but can be used to profile code running CUDA 9. You don’t get a timeline view, but you get many low level statistics about each individual kernel executed and can compare multiple runs (i.e. create a baseline).</p> <p><img alt="Nsight Compute" src="https://raw.githubusercontent.com/dmlc/web-data/master/mxnet/tutorials/python/profiler/profile_nsight_compute.png" /></p> diff --git a/date.txt b/date.txt deleted file mode 100644 index 04e0f7d..0000000 --- a/date.txt +++ /dev/null @@ -1 +0,0 @@ -Thu Jul 2 06:44:32 UTC 2020 diff --git a/feed.xml b/feed.xml index 487888c..a453885 100644 --- a/feed.xml +++ b/feed.xml @@ -1 +1 @@ -<?xml version="1.0" encoding="utf-8"?><feed xmlns="http://www.w3.org/2005/Atom" ><generator uri="https://jekyllrb.com/" version="4.0.0">Jekyll</generator><link href="https://mxnet.apache.org/feed.xml" rel="self" type="application/atom+xml" /><link href="https://mxnet.apache.org/" rel="alternate" type="text/html" /><updated>2020-07-02T06:32:45+00:00</updated><id>https://mxnet.apache.org/feed.xml</id><title type="html">Apache MXNet</title><subtitle>A flexible and efficient library for deep [...] \ No newline at end of file +<?xml version="1.0" encoding="utf-8"?><feed xmlns="http://www.w3.org/2005/Atom" ><generator uri="https://jekyllrb.com/" version="4.0.0">Jekyll</generator><link href="https://mxnet.apache.org/feed.xml" rel="self" type="application/atom+xml" /><link href="https://mxnet.apache.org/" rel="alternate" type="text/html" /><updated>2020-07-02T18:30:41+00:00</updated><id>https://mxnet.apache.org/feed.xml</id><title type="html">Apache MXNet</title><subtitle>A flexible and efficient library for deep [...] \ No newline at end of file