Embeddings Quickstart

Overview

Use the Gradient Embeddings API to easily generate embeddings to create your LLM's extended knowledge base.

🏕️ Set up your environment

If you have not used the Gradient API before, please follow the steps in CLI Quickstart or SDK Quickstart to set up your environment.

📦 Prepare your dataset

Your data should be in a JSONL file, where each line has the format:

{ "input": "<your-string>" }

🔢 Generate embeddings with the CLI

You can see the list of available embeddings models:

$ gradient embeddings list

View the available embeddings models here.

You can select an embeddings model and generate embeddings for your data with gradient embeddings generate <model-slug> <json-filepath>:

$ gradient embeddings generate bge-large ~/sample.jsonl

🔢 Generate embeddings with the SDKs

Here is an example of how to generate embeddings using the Python SDK:

from dotenv import load_dotenv

load_dotenv()
from gradientai import Gradient


def main() -> None:
    gradient = Gradient()

    embeddings_model = gradient.get_embeddings_model(slug="bge-large")

    generate_embeddings_response = embeddings_model.generate_embeddings(
        inputs=[
            "Multimodal brain MRI is the preferred method to evaluate for acute ischemic infarct and ideally should be obtained within 24 hours of symptom onset, and in most centers will follow a NCCT",
            "CTA has a higher sensitivity and positive predictive value than magnetic resonance angiography (MRA) for detection of intracranial stenosis and occlusion and is recommended over time-of-flight (without contrast) MRA",
            "Echocardiographic strain imaging has the advantage of detecting early cardiac involvement, even before thickened walls or symptoms are apparent",
        ],
    )

    for embedding in generate_embeddings_response.embeddings:
        print(f"generated embedding: {embedding.embedding}")

    gradient.close()


if __name__ == "__main__":
    main()

Here is an example to generate embeddings using the Typescript SDK:

import { Gradient } from "@gradientai/nodejs-sdk";

const main = async () => {
  const gradient = new Gradient({});

  const embeddingsModel = await gradient.getEmbeddingsModel({
    slug: "bge-large",
  });

  const { embeddings } = await embeddingsModel.generateEmbeddings({
    inputs: [
      "Multimodal brain MRI is the preferred method to evaluate for acute ischemic infarct and ideally should be obtained within 24 hours of symptom onset, and in most centers will follow a NCCT",
      "CTA has a higher sensitivity and positive predictive value than magnetic resonance angiography (MRA) for detection of intracranial stenosis and occlusion and is recommended over time-of-flight (without contrast) MRA",
      "Echocardiographic strain imaging has the advantage of detecting early cardiac involvement, even before thickened walls or symptoms are apparent",
    ],
  });

  for (const { embedding } of embeddings) {
    console.log(`created embedding: ${JSON.stringify(embedding, null, 2)}`);
  }
};

main()
  .catch((e) => {
    console.error(e);
    process.exit(1);
  })
  .finally(() => process.exit());