EWA MATEJSKA: Hi, everybody.

Thank you for joining us.

I'm Ewa Matejska, a technical program

manager on the TensorFlow team.

GAL OSHRI: I'm Gal Oshri, a product

manager on the TensorFlow team.

EWA MATEJSKA: So tell me, what's TensorBoard?

GAL OSHRI: So TensorBoard is TensorFlow's visualization tool

kit.

It is a popular tool for ML researchers and engineers

to understand their ML experiment results,

from tracking metrics to visualizing the model,

inspecting parameters, embeddings, and a lot more.

EWA MATEJSKA: How do I learn more about TensorBoard?

GAL OSHRI: So the best place to go is

to tensorflow.org/tensorboard, where we have a variety

of different tutorials you can easily follow to understand

what TensorBoard offers for you.

Each of these tutorials runs in the Google Colab,

so you can easily click, run Google Colab,

and then view that tutorial and run it yourself

to see the results and open TensorBoard directly

within the Colab.

EWA MATEJSKA: So are these available to try right now?

GAL OSHRI: Absolutely.

So if you go to the tensorflow.org/tensorboard

website, you can see all those tutorials, click on them,

and run them in Google Colab.

EWA MATEJSKA: Fantastic.

So what is new this year from last year?

GAL OSHRI: Absolutely.

So late last year, we launched tensorboard.dev.

We noticed a lot of people were taking screenshots

of their TensorBoard and sharing it in papers and GitHub repos

and blog posts and so on.

Now, the problem is that a screenshot

is not interactive and doesn't convey all the information.

We just saw that TensorBoard has so many capabilities,

but a screenshot just captures one part of that.

So we wanted to make it easy for people

to upload their TensorBoard logs and get a link that they can

then share with everyone and give them

the full experience of their TensorBoard.

EWA MATEJSKA: That sounds awesome.

Can you show me how to actually do that?

GAL OSHRI: Absolutely.

So if you go to TensorBoard.dev, you'll

learn a bit more about what TensorBoard.dev offers.

And you can click on the example Colab to learn how to get

started.

So over here, you can see that we

have a very simple model that we're

training on the MNIST data set.

And then later, we have the TensorBoard.dev upload command.

This takes the logs that I want to upload, as well

as, as of last week, I can include a name

and description to the experiment to give a little bit

more context.

If you open the TensorBoard.dev link,

you can immediately see, what was the Colab that

created this, or include a link to the GitHub repository

or the paper that you want to reference, or just

explain what is really interesting

about this TensorBoard.

So as soon as I start running this,

I can get a link to the TensorBoard.

And when I follow that link, I can open that TensorBoard

immediately, even if the experiment was still

in progress, and view those results.

I can now click on the top right Share button over here

to share it on various social media, or just copy the link

and include it wherever I want.

EWA MATEJSKA: Very cool.

I can't wait to share this with my mother.

So tell me, if there's one thing I need to take away

from this demo, what is it?

GAL OSHRI: So if you're not familiar with TensorBoard,

I suggest checking out tensorflow.org/tensorboard

to learn more about TensorBoard's capabilities

and how to use it.

If you are familiar with it and you

want to share your experiments, check out TensorBoard.dev

and follow the example Colab tutorial.

EWA MATEJSKA: Thank you for telling

me a bit more about TensorBoard and what's new.

And thank you for joining us.

GAL OSHRI: Thank you.

[MUSIC PLAYING]