Reference · How-to · ~8 min
How to sketch embeddings on Vertex AI
Last updated
Generate text embeddings on Google Cloud for search, RAG, or clustering pipelines.
Generate text embeddings on Google Cloud for search, RAG, or clustering pipelines.
#Steps
1. **Enable Vertex AI API** — set project, region, and service account with `aiplatform.user`
2. **Pick model** — e.g. `text-embedding-005` (check current GA model list)
3. **Batch embed** — send list of strings; respect token limits per request
4. **Store vectors** — Vertex Vector Search, BigQuery vector columns, or Chroma locally
5. **Query loop** — embed question → nearest neighbors → pass text to Gemini
6. **Monitor cost** — batch offline jobs; cache repeated strings
#Sketch
from vertexai.language_models import TextEmbeddingModel
import vertexai
vertexai.init(project="my-project", location="us-central1")
model = TextEmbeddingModel.from_pretrained("text-embedding-005")
embeddings = model.get_embeddings(["refund policy summary", "shipping FAQ"])
vectors = [e.values for e in embeddings]#Watch out
Model names and dimensions change — pin version in config and re-embed when upgrading.