Sampling
The sampling tab is where you can control what samples you want to create and when.
On the main tab you can do the following:
- Sample after X Y. In this, x is an integer and y can be any of the following Epoch/Step/Second/Minute/Hour/Never/Always. Note that Never and Always are special and ignore the integer input.
- Choose the Picture format. The default is JPG. Also available is PNG
- Sample now. When you press this button, OneTrainer will sample at the next available opportunity
- Manual sample. Pressing this button will open the manual sample window, see more info below.
- Non-EMA sampling. Enabling this toggle will allow generation of EMA and non-EMA sampling when EMA is used.
- Samples to Tensorboard. Enabling this toggle will allow samples to be loaded as part of your Tensorboard run data.
- The dropdown menu will show your sample configurations that have been created
- The add config button will allow you to create a new sample configuration file
- The add sample button will create a new sample in the space below
- Note: One way to easily duplicate sample from one config to another is to go to your training_samples folder and find the .json file you want to copy. Make a copy of it and rename it to the config name you would like. You can then load this new config in OneTrainer and modify it. Or it can be modified in your text editor of choice.
- In this area, the sample prompt(s) will appear and you can do some high level modification of them here.
- To see all options, you must press the ... button to the right of the sample.
- You can use the buttons in front of the sample to easily delete or duplicate the sample.
- The toggle in the front of a sample controls if the sample is enabled or not. Note: It is not possible to edit using the ... button samples that are not enabled.
- Note - This tab can be updated during training and will update your samples and frequency when you make changes. This can lead to unattended consequences if you start to setup your next runs samples, and your current training starts a sample generation run.