Vertex AI GoogleAuthError in nodeJS

Question

I have a web server with one endpoint deployed on Heroku. Server purpose is to interact with Vertex AI gemini-pro LLM. Localy everything wors perfect because I am using a gCloud CLI auth but I want to implement a service account auth.

Here is the code sample of model init:

import { VertexAI } from "@google-cloud/vertexai";
import { GoogleAuth } from "google-auth-library";
import dotenv from "dotenv";

dotenv.config();

const gAuth = new GoogleAuth({
  credentials: {
    client_email: process.env.CLIENT_EMAIL,
    private_key: process.env.PRIVATE_KEY,
  },
});

const authClient = await gAuth.getClient();

const vertex_ai = new VertexAI({
  project: process.env.PROJECT_ID,
  location: process.env.LOCATION,
  googleAuth: authClient,
});

const model = "gemini-pro";
const generativeModel = vertex_ai.preview.getGenerativeModel({
  model: model,
  generation_config: {
    max_output_tokens: 8192,
    temperature: 0.8,
    top_p: 0.8,
    top_k: 5,
  },
});

export default generativeModel;

I checked the service acc role - it's "Editor" so the problem is in model init. Every time when trying to reach and endpoint I get the error:

2024-02-05T21:13:23.977555+00:00 app[web.1]: GoogleAuthError: 
2024-02-05T21:13:23.977557+00:00 app[web.1]: Unable to authenticate your request        
2024-02-05T21:13:23.977576+00:00 app[web.1]: Depending on your run time environment, you can get authentication by        
2024-02-05T21:13:23.977576+00:00 app[web.1]: - if in local instance or cloud shell: `!gcloud auth login`        
2024-02-05T21:13:23.977576+00:00 app[web.1]: - if in Colab:        
2024-02-05T21:13:23.977577+00:00 app[web.1]:     -`from google.colab import auth`        
2024-02-05T21:13:23.977577+00:00 app[web.1]:     -`auth.authenticate_user()`        
2024-02-05T21:13:23.977578+00:00 app[web.1]: - if in service account or other: please follow guidance in https://cloud.google.com/docs/authentication
2024-02-05T21:13:23.977579+00:00 app[web.1]: Error: Could not load the default credentials. Browse to https://cloud.google.com/docs/authentication/getting-started for more information.

UPD: I found the solution in migrating to using API key:

import { GoogleGenerativeAI } from "@google/generative-ai";
import dotenv from "dotenv";

dotenv.config();

const genAI = new GoogleGenerativeAI(process.env.API_KEY);

const model = "gemini-pro";
const generativeModel = genAI.getGenerativeModel({
  model: model,
  generation_config: {
    max_output_tokens: 8192,
    temperature: 0.8,
    top_p: 0.8,
    top_k: 5,
  },
});

export default generativeModel;

Answer 1

There are a couple of issues, I think, with how you're setting up the various authentication elements. Do you have a reference for where you were getting these attributes?

The biggest is that the VertexAI constructor doesn't have a googleAuth attribute in its initializing object which is of type VertexInit. But it does have a googleAuthOptions attribute that is of type GoogleAuthOptions.

So I think your code should look something like

const authOptions = {
  credentials: {
    client_email: process.env.CLIENT_EMAIL,
    private_key: process.env.PRIVATE_KEY,
  }
}

const vertex_ai = new VertexAI({
  project: process.env.PROJECT_ID,
  location: process.env.LOCATION,
  googleAuthOptions: authOptions,
});

It is also worth checking to make sure the various values that are coming from the environment are actually populated with the values you think they are.