Galileo Datasets allow you to create, manage, and version collections of data for testing and evaluating your LLM applications.

​
Creating and Using Datasets

You can create and use datasets for experimentation:

import { createDataset, getDataset } from "galileo";

// Create a dataset from a file path
const datasetFromFile = await createDataset("/path/to/dataset.csv", "my-dataset-name");

// Create a dataset from an array of objects
const datasetFromArray = await createDataset(
  [
    { input: "Spain", output: "Europe" },
    { input: "Japan", output: "Asia" },
  ],
  "countries",
);

// Create a dataset from a dictionary of arrays
const datasetFromDict = await createDataset(
  {
    input: ["Spain", "Japan"],
    output: ["Europe", "Asia"],
  },
  "countries-from-dict",
);

// Get a dataset by name
const dataset = await getDataset({ name: "countries" });

​
Using Datasets in Experiments

Datasets can be used in experiments to evaluate your LLM applications:

import { GalileoScorers, runExperiment } from "galileo";
import { OpenAI } from "openai";

async function runDatasetExperiment() {
  const openai = new OpenAI({ apiKey: process.env.OPENAI_API_KEY });

  const runner = async (input: any) => {
    const result = await openai.chat.completions.create({
      model: "gpt-4o",
      messages: [{ content: `Which continent does the following country belong to ${input["input"]}!`, role: "user" }],
    });
    return result;
  };

  await runExperiment({
    name: `Test Experiment`,
    datasetName: "countries",
    function: runner,
    metrics: [GalileoScorers.Correctness],
    projectName: "my-project",
  });
}

// Run the experiment
runDatasetExperiment();

​
Using Custom Datasets

You can also use locally generated datasets for experiments:

import { GalileoScorers, runExperiment } from "galileo";
import { OpenAI } from "openai";

async function runCustomDatasetExperiment() {
  const openai = new OpenAI({ apiKey: process.env.OPENAI_API_KEY });

  const dataset = [
    {
      input: "Spain",
      output: "Europe",
    },
    {
      input: "Japan",
      output: "Asia",
    },
  ];

  const runner = async (input: any) => {
    const result = await openai.chat.completions.create({
      model: "gpt-4o",
      messages: [{ content: `Which continent does the following country belong to ${input["input"]}!`, role: "user" }],
    });
    return result;
  };

  await runExperiment({
    name: `Test Experiment`,
    dataset: dataset,
    function: runner,
    metrics: [GalileoScorers.InputToxicity, GalileoScorers.OutputToxicity],
    projectName: "my-project",
  });
}

// Run the experiment
runCustomDatasetExperiment();