Add cifar10 inference with resnet20 model

[code-perspective]

No description provided.

[shaih]

So we are treating them as two different workloads entirely? Why did we decide to go this way rather than having one big "ML inference" and specifying the model?

[code-perspective]

So we are treating them as two different workloads entirely? Why did we decide to go this way rather than having one big "ML inference" and specifying the model?

The underlying repository is the same (for MNIST and CIFAR-10) but the website has two different pages. Im open to having one large "ML Inference" workload but then again BERT is a different repository. we could go with

1. Separate webpage for MNIST, CIFAR10, BERT listed separately on the main page. (as it is currently)
2. Separate webpage for MNIST, CIFAR10, BERT listed under ML Inference umbrella even if the repositories are different.
3. Treat BERT (and perhaps face recognition) differently from ML Inference. This seemed inconsistent to me.

Curious to hear if you think of any other options.

[andreea-alexandru]

What about a page per repository? CIFAR-10 and MNIST are in the same repository, with the same measurement-styles. We can have one table per dataset on the same page (we'll worry about long scrolling when we get too many submissions). BERT and face recognition are different repositories, so they should each have their own page.

[code-perspective]

What about a page per repository? CIFAR-10 and MNIST are in the same repository, with the same measurement-styles. We can have one table per dataset on the same page (we'll worry about long scrolling when we get too many submissions). BERT and face recognition are different repositories, so they should each have their own page.

What would we categorize CIFAR-10 and MNIST as? ML inference seems a bit inconsistent since BERT and face recognition is also ML Inference.

Maybe Image classification?