[Feature Request] Can we get access to text-multilingual-embedding models

[Champkinz]

Is your feature request related to a problem? Please describe.
I have a requirement to embed non English text, but at the moment i can only access textEmbeddingGecko through,
import { textEmbeddingGecko, vertexAI } from '@genkit-ai/vertexai';

Describe the solution you'd like
A way to use a multilingual embedding model through genkit

Describe alternatives you've considered
Using the Vertex python SDK

I don't want mix too many SDKs, want to keep it concise so we can maintain our solution without a hassle.

[Champkinz]

So i did some digging and found this comment

/*
// @TODO(huangjeff): Fix multilingual text embedding gecko
// For some reason this model returns 404 but it exists in the reference docs:
// https://cloud.google.com/vertex-ai/generative-ai/docs/embeddings/get-text-embeddings

export const textEmbeddingGeckoMultilingual001 = embedderRef({
  name: 'vertexai/textembedding-gecko-multilingual@001',
  configSchema: TextEmbeddingGeckoConfigSchema,
  info: {
    dimensions: 768,
    label: 'Vertex AI - Multilingual Text Embedding Gecko',
    supports: {
      input: ['text'],
    },
  },
});
*/

export const textEmbeddingGecko = textEmbeddingGecko003;

export const SUPPORTED_EMBEDDER_MODELS: Record<string, EmbedderReference> = {
  'textembedding-gecko@003': textEmbeddingGecko003,
  'textembedding-gecko@002': textEmbeddingGecko002,
  'textembedding-gecko@001': textEmbeddingGecko001,
  //'textembeddding-gecko-multilingual@001': textEmbeddingGeckoMultilingual001,
};

I'll be working on fixing this or try the defineEmbedder method

if successful, I can raise a PR for this.

[Champkinz]

I FIGURED IT OUT 🙌

The issue here is
//'textembeddding-gecko-multilingual@001': textEmbeddingGeckoMultilingual001,

has an extra 'd' in the model,

remove that extra d and it starts working, i have tested this locally and i can confirm its working

 const SUPPORTED_EMBEDDER_MODELS = {
  "textembedding-gecko@003": textEmbeddingGecko003,
  "textembedding-gecko@002": textEmbeddingGecko002,
  "textembedding-gecko@001": textEmbeddingGecko001,
  'textembedding-gecko-multilingual@001': textEmbeddingGeckoMultilingual001,
};

The 404 error is because the end point was calling with the wrong embedder name

https://<selected_region>-aiplatform.googleapis.com/v1/projects/<project_id>/locations/<selected_region>/publishers/google/models/textembeddding-gecko-multilingual@001:predict

instead of

https://<selected_region>-aiplatform.googleapis.com/v1/projects/<project_id>/locations/<selected_region>/publishers/google/models/textembedding-gecko-multilingual@001:predict

[pavan168]

+1

[pavelgj]

These were added in #340 and released in 0.5.3.