EmbeddingGemma 2 开发者指南

英文标题：EmbeddingGemma 2: The Developer Guide

EmbeddingGemma 2 是一款采用 Apache 2.0 许可证发布的紧凑型开源多模态嵌入模型，可将文本、代码、图像、视频和音频映射到统一的 768 维向量空间。该模型基于 Gemma 4，开发者可按需加载编码器，参数规模可从仅支持文本和代码的 2.7 亿扩展至全模态的 7.4 亿。通过 Matryoshka Representation Learning（MRL），向量可从 768 维截断至 128 维，以显著降低向量数据库存储成本。对于 RAG、代码搜索、跨模态检索和端侧索引等场景，sentence-transformers 库提供了直接的模型加载、任务提示、媒体输入和向量截断支持。

11.7 KB 54.9 KB 674.8 KB 204.4 KB 102.2 KB 105.2 KB 730.2 KB 47.0 KB 351.9 KB 46.3 KB 1.0 MB

pip install -U sentence-transformers[image,audio,video] transformers

from sentence_transformers import SentenceTransformer

# 完整模型：所有模态（7.4 亿参数）
MODEL_ID = "google/embeddinggemma-2"
model = SentenceTransformer(MODEL_ID)

# 仅文本（2.7 亿参数）
text_only_model = SentenceTransformer(
    MODEL_ID,
    config_kwargs={"vision_config": None, "audio_config": None},
)

# 文本、图像和视频（4.4 亿参数）
text_image_model = SentenceTransformer(
    MODEL_ID,
    config_kwargs={"audio_config": None},
)

# 文本和音频（5.7 亿参数）
text_audio_model = SentenceTransformer(
    MODEL_ID,
    config_kwargs={"vision_config": None},
)

query = "What causes the northern lights?"
document = "The northern lights are caused by charged particles from the sun.." # truncated

# 使用 `prompt_name` 生成嵌入
query_emb = model.encode(query, prompt_name="SearchQuery")
doc_emb = model.encode(document, prompt_name="Document")

print(model.similarity(query_emb, doc_emb))

# 跨模态搜索：一个文本查询与一张照片和一段录音进行匹配
image_emb = model.encode({"image": "sunset_beach.jpg"})
audio_emb = model.encode({"audio": "ocean_waves.wav"})
query_emb = model.encode("ocean waves at sunset", prompt_name="SearchQuery")

print(model.similarity(query_emb, image_emb))
print(model.similarity(query_emb, audio_emb))

# 交错输入：为包含文本、照片和视频的商品列表生成一个嵌入
listing_emb = model.encode({
    "text": "Waterproof trail shoe. <|image|> Grip test on wet rock: <|video|>",
    "image": "trail_shoe.jpg",
    "video": "grip_test.mp4",
})
query_emb = model.encode("waterproof trail shoes", prompt_name="SearchQuery")

print(model.similarity(query_emb, listing_emb))

# 截断查询向量
query_emb = model.encode(
    query,
    prompt_name="SearchQuery",
    truncate_dim=256,
    normalize_embeddings=True,
)

pip install -U sentence-transformers[image,audio,video] transformers

from sentence_transformers import SentenceTransformer

# Full model: all modalities (740M parameters)
MODEL_ID = "google/embeddinggemma-2"
model = SentenceTransformer(MODEL_ID)

# Text only (270M parameters)
text_only_model = SentenceTransformer(
    MODEL_ID,
    config_kwargs={"vision_config": None, "audio_config": None},
)

# Text, images, and video (440M parameters)
text_image_model = SentenceTransformer(
    MODEL_ID,
    config_kwargs={"audio_config": None},
)

# Text and audio (570M parameters)
text_audio_model = SentenceTransformer(
    MODEL_ID,
    config_kwargs={"vision_config": None},
)

query = "What causes the northern lights?"
document = "The northern lights are caused by charged particles from the sun.." # truncated

# Embed using `prompt_name`
query_emb = model.encode(query, prompt_name="SearchQuery")
doc_emb = model.encode(document, prompt_name="Document")

print(model.similarity(query_emb, doc_emb))

# Cross-modal search: one text query against a photo and a sound recording
image_emb = model.encode({"image": "sunset_beach.jpg"})
audio_emb = model.encode({"audio": "ocean_waves.wav"})
query_emb = model.encode("ocean waves at sunset", prompt_name="SearchQuery")

print(model.similarity(query_emb, image_emb))
print(model.similarity(query_emb, audio_emb))

# Interleaved: one embedding for a product listing with text, photo, and video
listing_emb = model.encode({
    "text": "Waterproof trail shoe. <|image|> Grip test on wet rock: <|video|>",
    "image": "trail_shoe.jpg",
    "video": "grip_test.mp4",
})
query_emb = model.encode("waterproof trail shoes", prompt_name="SearchQuery")

print(model.similarity(query_emb, listing_emb))

# Truncate the query
query_emb = model.encode(
    query,
    prompt_name="SearchQuery",
    truncate_dim=256,
    normalize_embeddings=True,
)