通过 Google 的 google-genai 包，LlamaIndex 提供了 GoogleGenAIEmbedding 类，允许你使用最新的 gemini-embedding-2-preview 模型，通过 Gemini 和 Vertex AI 两个 API 的 Google GenAI 模型嵌入文本。

如果你在 Colab 中打开这个 Notebook，可能需要先安装 LlamaIndex 🦙。

%pip install llama-index-embeddings-google-genai

import os

os.environ["GOOGLE_API_KEY"] = "..."

设置

GoogleGenAIEmbedding 是围绕 google-genai 包的封装，这意味着它开箱即支持 Gemini 和 Vertex AI 两个 API。

你可以直接传入 api_key，或传入 vertexai_config 以使用 Vertex AI API。

其他选项包括 embed_batch_size、model_name 和 embedding_config。

from llama_index.embeddings.google_genai import GoogleGenAIEmbedding
from google.genai.types import EmbedContentConfig

embed_model = GoogleGenAIEmbedding(
    model_name="gemini-embedding-2-preview",
    embed_batch_size=100,
    # can pass in the api key directly
    # api_key="...",
    # or pass in a vertexai_config
    # vertexai_config={
    #     "project": "...",
    #     "location": "...",
    # }
    # can also pass in an embedding_config
    # embedding_config=EmbedContentConfig(...)
)

用量

同步

embeddings = embed_model.get_text_embedding("Google Gemini Embeddings.")
print(embeddings[:5])
print(f"Dimension of embeddings: {len(embeddings)}")

[0.031099992, 0.02192731, -0.06523498, 0.016788177, 0.0392835]
Dimension of embeddings: 768

embeddings = embed_model.get_query_embedding("Query Google Gemini Embeddings.")
print(embeddings[:5])
print(f"Dimension of embeddings: {len(embeddings)}")

[0.022199392, 0.03671178, -0.06874573, 0.02195774, 0.05475164]
Dimension of embeddings: 768

embeddings = embed_model.get_text_embedding_batch(
    [
        "Google Gemini Embeddings.",
        "Google is awesome.",
        "Llamaindex is awesome.",
    ]
)
print(f"Got {len(embeddings)} embeddings")
print(f"Dimension of embeddings: {len(embeddings[0])}")

Got 3 embeddings
Dimension of embeddings: 768

异步

embeddings = await embed_model.aget_text_embedding("Google Gemini Embeddings.")
print(embeddings[:5])
print(f"Dimension of embeddings: {len(embeddings)}")

[0.031099992, 0.02192731, -0.06523498, 0.016788177, 0.0392835]
Dimension of embeddings: 768

embeddings = await embed_model.aget_query_embedding(
    "Query Google Gemini Embeddings."
)
print(embeddings[:5])
print(f"Dimension of embeddings: {len(embeddings)}")

[0.022199392, 0.03671178, -0.06874573, 0.02195774, 0.05475164]
Dimension of embeddings: 768

embeddings = await embed_model.aget_text_embedding_batch(
    [
        "Google Gemini Embeddings.",
        "Google is awesome.",
        "Llamaindex is awesome.",
    ]
)
print(f"Got {len(embeddings)} embeddings")
print(f"Dimension of embeddings: {len(embeddings[0])}")

Got 3 embeddings
Dimension of embeddings: 768