We use a simple user interface to run experiments on customized cloud platform.
Here’s a step-by-step guide to create and run an experiment:
Let's create an experiment at first. Click "Create New" to create a new experiment or "History" to use/modify a previous experiment
Set up the environment for your experiment 1) Enter "Name" as the name of the experiment and "Description" to describe the details of the experiment 2) Choose "Image" which includes the environment and basic setup-package 3) Choose the core number of "CPU", size of "RAM" and type of "GPU"
You can choose on-demand or spot instance in "Instance type" (Spot instance will cost less but use an unstable instance)
Currently, only certain combinations of CPU, RAM and GPU will lead to a successful experiment:
| CPU | RAM | GPU |
|---|---|---|
| 2 | 2, 4, 8 | None |
| 4 | 16, 61 | None, K80 |
| 8 | 32, 61 | None, V100 |
| 16 | 122 | M60 |
Upload your model function with a trainer to find the optimal weights and hyperparameters, or you can only design your own model and use our trainer
You can upload / delete function, model and requirements here.
User Function/Model will be saved in the File Manager.
Our **Magic Object** will automatically load function or model arguments into Function Config. If you just have a model without trainer, our Trainer Config can set hyperparameters for you. There are some basic settings in Trialer Config for the experiment to follow.