Experiments

Experiments help you learn how your latest Pipe config will affect LLM response by running it against your previous five generate requests. They are only available for Generate Pipes.

Experiments contains a Previous Completions column that are results of your past five runs.

When you run the Experiments, Pipe executes the most recent Pipe configuration against the last generate request prompts and shows the outcomes in the New Completions column.

One example can be changing Pipe’s LLM model to gemma-7b-it from gpt-4-turbo-preview to check how the response will look like.