LlamaIndex Embeddings Integration: Vertex

Implements Vertex AI Embeddings Models:

Model	Release Date
textembedding-gecko@003	December 12, 2023
textembedding-gecko@002	November 2, 2023
textembedding-gecko-multilingual@001	November 2, 2023
textembedding-gecko@001	June 7, 2023
multimodalembedding

Note: Currently Vertex AI does not support async on multimodalembedding. Otherwise, VertexTextEmbedding supports async interface.

New Features

Flexible Credential Handling:
- Credential Management: Supports both direct credentials and service account info for secure API access.
Model Name Handling in Embedding Requests:
- The _get_embedding_request function now accepts the model_name parameter, allowing it to manage models that do not support the task_type parameter, like textembedding-gecko@001.

Example Usage

python

from google.oauth2 import service_account
from llama_index.embeddings.vertex import VertexTextEmbedding

credentials = service_account.Credentials.from_service_account_file(
    "path/to/your/service-account.json"
)

embedding = VertexTextEmbedding(
    model_name="textembedding-gecko@003",
    project="your-project-id",
    location="your-region",
    credentials=credentials,
)

Alternatively, you can directly pass the required service account parameters:

python

from llama_index.embeddings.vertex import VertexTextEmbedding

embedding = VertexTextEmbedding(
    model_name="textembedding-gecko@003",
    project="your-project-id",
    location="your-region",
    client_email="your-service-account-email",
    token_uri="your-token-uri",
    private_key_id="your-private-key-id",
    private_key="your-private-key",
)