Implementation:Vibrantlabsai Ragas GoogleEmbeddings

Knowledge Sources	Vibrantlabsai_Ragas
Domains	Embeddings, Google AI, Vertex AI
Last Updated	2026-02-12 00:00 GMT

Overview

Provides a unified Google embeddings implementation supporting both Vertex AI and Google AI (Gemini) through the new google-genai SDK and the legacy google-generativeai SDK.

Description

The GoogleEmbeddings class is a concrete implementation of BaseRagasEmbedding that provides text embedding capabilities via Google's embedding services. It supports three distinct backends:

• Google AI (Gemini) via new SDK (google-genai) -- The recommended approach. Uses a genai.Client() instance with the client.models.embed_content() interface. Supports batch embedding natively.

• Google AI (Gemini) via old SDK (google-generativeai) -- The deprecated approach (support ends August 2025). Uses the genai.embed_content() module-level function. Does not support native batch processing; texts are embedded individually.

• Vertex AI -- Enterprise Google Cloud embedding via google-cloud-aiplatform. Uses TextEmbeddingModel.from_pretrained() and supports batch embedding.

The class performs intelligent client resolution in its constructor. When no client is provided, it attempts auto-import of the new SDK first, then falls back to the old SDK. When a client is provided, it inspects the client's module path and class structure to determine which SDK interface to use. The _use_new_sdk flag tracks the resolved SDK path at runtime.

Async operations are implemented by delegating synchronous calls to a thread pool via run_sync_in_async, since Google's SDKs do not provide native async embedding support.

Usage

Import this class when you need to embed text using Google's embedding models, either through Google AI (Gemini) or Vertex AI. It is also used internally by the embedding_factory when the provider is set to "google".

Code Reference

Source Location

• Repository: Vibrantlabsai_Ragas
• File: src/ragas/embeddings/google_provider.py

Signature

class GoogleEmbeddings(BaseRagasEmbedding):
    PROVIDER_NAME = "google"
    REQUIRES_CLIENT = False
    DEFAULT_MODEL = "gemini-embedding-001"

    def __init__(
        self,
        client: t.Optional[t.Any] = None,
        model: str = "gemini-embedding-001",
        use_vertex: bool = False,
        project_id: t.Optional[str] = None,
        location: t.Optional[str] = "us-central1",
        cache: t.Optional[CacheInterface] = None,
        **kwargs: t.Any,
    ): ...

    def embed_text(self, text: str, **kwargs: t.Any) -> t.List[float]: ...
    async def aembed_text(self, text: str, **kwargs: t.Any) -> t.List[float]: ...
    def embed_texts(self, texts: t.List[str], **kwargs: t.Any) -> t.List[t.List[float]]: ...
    async def aembed_texts(self, texts: t.List[str], **kwargs: t.Any) -> t.List[t.List[float]]: ...

Import

from ragas.embeddings.google_provider import GoogleEmbeddings

I/O Contract

Constructor Inputs

Name	Type	Required	Description
client	Any or None	No	Pre-initialized client: genai.Client() for new SDK, genai module for old SDK, or Vertex AI client. If None, auto-imports.
model	str	No (default "gemini-embedding-001")	The embedding model name (e.g., "gemini-embedding-001", "text-embedding-004")
use_vertex	bool	No (default False)	Whether to use Vertex AI instead of Google AI (Gemini)
project_id	str or None	No	Google Cloud project ID (used with Vertex AI)
location	str or None	No (default "us-central1")	Google Cloud region (used with Vertex AI)
cache	CacheInterface or None	No	Optional cache backend for caching embedding results
**kwargs	Any	No	Additional provider-specific arguments passed to the embedding API calls

embed_text Inputs

Name	Type	Required	Description
text	str	Yes	The text string to embed
**kwargs	Any	No	Additional arguments merged with constructor kwargs and passed to the embedding API

embed_text Outputs

Name	Type	Description
return	List[float]	A list of floats representing the embedding vector

embed_texts Inputs

Name	Type	Required	Description
texts	List[str]	Yes	List of text strings to embed
**kwargs	Any	No	Additional arguments for the embedding API calls

embed_texts Outputs

Name	Type	Description
return	List[List[float]]	A list of embedding vectors, one per input text

Usage Examples

New SDK (Recommended)

from google import genai
from ragas.embeddings.google_provider import GoogleEmbeddings

client = genai.Client(api_key="your-api-key")
embeddings = GoogleEmbeddings(client=client, model="gemini-embedding-001")

# Single text
vector = embeddings.embed_text("What is machine learning?")

# Batch texts
vectors = embeddings.embed_texts(["Text one", "Text two", "Text three"])

Old SDK (Deprecated)

import google.generativeai as genai
from ragas.embeddings.google_provider import GoogleEmbeddings

genai.configure(api_key="your-api-key")
embeddings = GoogleEmbeddings(client=genai, model="text-embedding-004")

vector = embeddings.embed_text("Hello world")

Auto-Import (No Client)

from ragas.embeddings.google_provider import GoogleEmbeddings

# Tries new SDK first, falls back to old SDK
embeddings = GoogleEmbeddings(model="text-embedding-004")
vector = embeddings.embed_text("Hello world")

Vertex AI

from ragas.embeddings.google_provider import GoogleEmbeddings

embeddings = GoogleEmbeddings(
    client=vertex_client,
    model="textembedding-gecko@003",
    use_vertex=True,
    project_id="my-gcp-project",
    location="us-central1",
)
vector = embeddings.embed_text("Embed this text via Vertex AI")

Via Factory

from ragas.embeddings.base import embedding_factory
from google import genai

client = genai.Client(api_key="your-api-key")
embedder = embedding_factory("google", model="gemini-embedding-001", client=client)
vector = embedder.embed_text("Factory-created embedding")

Related Pages

• Vibrantlabsai_Ragas_BaseRagasEmbeddings -- Parent abstract base class providing the embedding interface
• google-genai -- Recommended Google AI SDK
• google-generativeai -- Deprecated Google AI SDK (fallback)
• google-cloud-aiplatform -- Required for Vertex AI mode