Hi there!
I鈥檇 like to run one-off / in-memory evaluations without persisting the datasets. I am using the Typescript client. Two related questions:
1 Ad-hoc dataset in runExperiment
Is there a supported way to supply the dataset directly to runExperiment like this?
const experiment = await runExperiment({
dataset: {
name: "qa-capital-france",
description: "Single-prompt dataset for capital city QA",
examples: [
{
input: { prompt: "What is the capital of France?" },
output: { text: "Paris" },
metadata: {},
},
],
},
task,
evaluators
});
2. Overwriting an existing dataset
If ad-hoc isn鈥檛 possible, I can call createDataset, but repeated runs fail with:
Dataset with the same name already exists: name='qa-capital-france'
Is there a way to overwrite / upsert so that I can recreate the datasets without bumping the name?