Number of Generations

When you call the Generate endpoint, you have the option to generate multiple generations in a single call. This is done by setting the num_generations parameter.

The model’s outputs will vary depending on the generation settings you have specified, such as temperature, top-k, and top-p.

Each generation comes with its set of likelihood values, which consists of:

• The likelihood of each generated token.
• The average likelihood of all generated tokens.

Example

This example uses the input: “This curved gaming monitor delivers ...”

The output generated with a maximum token set of 4 and sorted by average token likelihood are:

You can use these outputs in a number of ways, by selecting the one with the highest likelihood as the final output, for example, or by presenting them as options in your application.