Comparing variants

Finding the optimal variant (with good enough quality and low enough price) is the main goal of the tool. Toloka LLM shows all the created variants in one right-side column.

The first variant appears automatically in the History right-side area when you add a new class.

You can choose a metric to compare the resulting variants with each other. For a metric to be displayed you need to have a labeled dataset and run a model.

Changing prompt invalidates quality measurement for the variant you are currently editing. Changing dataset invalidates quality measurement for all variants.

After you modify the prompt or change the dataset, click Run test Run test to create a new variant.

The created variant will be displayed in the History area. You can compare the variant accuracy and price and select the best variants for your model.

When you have a variant with the best accuracy to price ratio (say, 85% accuracy to $4 per 1K items ratio is better than 88% accuracy to $9 per 1K items even though the accuracy is higher in the second variant), or the variant that fits you both in accuracy and price, deploy the project and use it with real datasets.

Variant actions

The following actions are available for your variants:

  • Sort the variants by accuracy or price using the Sort by option.

  • Add a variant to favorite clicking Favorite. It will turn to Favorited and add the variant to the comparison in the table below. To remove the variant from the comparison, click Favorited once again.

  • Delete a variant clicking Delete variant.

    Note

    When there is only one variant it can't be deleted.

Next steps

Last updated: September 11, 2023

Introduction
Toloka LLMRegistering and signing inAdding OpenAI API keyTerminology
Working with Toloka LLM
Workflow overview
Working with datasets
Problem definition
Iterating on quality
Creating endpoint