Skip to main content

Interacting with Experiments

Sorting

You can sort Experiments run on a dataset using the grey sorting bar at the top of your Experiments.

Sorting Bar

The different sorting options are:

  • Date: Shows the latest Experiments first
  • Mine: Shows your own Experiments first
  • Upvote: Shows the Experiment with the most upvotes first
  • acc-flops, acc-gCO2e, vgap-acc, vgap: Each shows the Experiment with the highest respective score first

Pausing and Stopping

You can pause or stop experiments that have not finished already. This includes experiments that just started and are waiting for the first results to come in.

  • Pausing an Experiment puts the training on hold. Paused Experiments can be resumed anytime by pressing on the "Start Experiment" button next to the "Pause Experiment" button.
  • Stopping an Experiment terminates the training. Stopped Experiments can not be resumed again.

You can pause, stop and resume your experiments here:

Pausing Experiments

Running Inference on Experiments

In order to evaluate your model performance you can submit an Experiment at a specific cycle here:

Submit Model

It takes sometime for the results to show on the experiments tab.

note

You can submit an experiment only once that cycle data is available on the experiment board

Downloading and archiving

You can download the model, weights and training plan of your or your teammates experiments using the "Download Model, Weights, Trainingplan" button.

To archive one of your experiments, click on the "Archive Experiment" button.

Both can be found here:

Download and Archive

Upvoting

You can upvote experiments to show your appreciation for them. The Upvote button is at the top left of an experiment:

Upvote

Confusion Matrix

The confusion matrix of an experiment can be found here:

Confusion Matrix

Training Plan

The Training Plan of an experiment can be found here:

Training Plan

Discussion

You can use the discussion feature to connect with your peers and share your thoughts on their experiments. It can be found here:

Disxussion

Model Code

The code for the model used in an experiment can be found here:

Model Code