PingKai Logo下载
搜索

Hugging Face 向量嵌入

本文档介绍如何在 TiDB Cloud 中结合 Auto Embedding 使用 Hugging Face 向量嵌入模型，通过文本查询实现语义搜索。

注意

Auto Embedding 仅适用于托管在 AWS 上的 TiDB Cloud Starter 实例。

可用模型

如果你自带 Hugging Face Inference API 密钥（BYOK），则可以通过 huggingface/ 前缀使用 Hugging Face 模型。

为方便起见，以下章节以几个流行模型为例。完整可用模型列表请参见 Hugging Face models。请注意，并非所有模型都可通过 Hugging Face Inference API 获取，或都能稳定运行。

multilingual-e5-large

  • 名称：huggingface/intfloat/multilingual-e5-large
  • 维度：1024
  • 距离度量：Cosine，L2
  • 价格：由 Hugging Face 收费
  • TiDB Cloud 托管：❌
  • Bring Your Own Key（BYOK，由用户自行提供 API 密钥）：✅
  • 项目主页：https://huggingface.co/intfloat/multilingual-e5-large

示例：

SET @@GLOBAL.TIDB_EXP_EMBED_HUGGINGFACE_API_KEY = 'your-huggingface-api-key-here';

CREATE TABLE sample (
  `id`        INT,
  `content`   TEXT,
  `embedding` VECTOR(1024) GENERATED ALWAYS AS (EMBED_TEXT(
                "huggingface/intfloat/multilingual-e5-large",
                `content`
              )) STORED
);


INSERT INTO sample
    (`id`, `content`)
VALUES
    (1, "Java: Object-oriented language for cross-platform development."),
    (2, "Java coffee: Bold Indonesian beans with low acidity."),
    (3, "Java island: Densely populated, home to Jakarta."),
    (4, "Java's syntax is used in Android apps."),
    (5, "Dark roast Java beans enhance espresso blends.");


SELECT `id`, `content` FROM sample
ORDER BY
  VEC_EMBED_COSINE_DISTANCE(
    embedding,
    "How to start learning Java programming?"
  )
LIMIT 2;

bge-m3

  • 名称：huggingface/BAAI/bge-m3
  • 维度：1024
  • 距离度量：Cosine，L2
  • 价格：由 Hugging Face 收费
  • TiDB Cloud 托管：❌
  • Bring Your Own Key（BYOK，由用户自行提供 API 密钥）：✅
  • 项目主页：https://huggingface.co/BAAI/bge-m3
SET @@GLOBAL.TIDB_EXP_EMBED_HUGGINGFACE_API_KEY = 'your-huggingface-api-key-here';

CREATE TABLE sample (
  `id`        INT,
  `content`   TEXT,
  `embedding` VECTOR(1024) GENERATED ALWAYS AS (EMBED_TEXT(
                "huggingface/BAAI/bge-m3",
                `content`
              )) STORED
);


INSERT INTO sample
    (`id`, `content`)
VALUES
    (1, "Java: Object-oriented language for cross-platform development."),
    (2, "Java coffee: Bold Indonesian beans with low acidity."),
    (3, "Java island: Densely populated, home to Jakarta."),
    (4, "Java's syntax is used in Android apps."),
    (5, "Dark roast Java beans enhance espresso blends.");


SELECT `id`, `content` FROM sample
ORDER BY
  VEC_EMBED_COSINE_DISTANCE(
    embedding,
    "How to start learning Java programming?"
  )
LIMIT 2;

all-MiniLM-L6-v2

  • 名称：huggingface/sentence-transformers/all-MiniLM-L6-v2
  • 维度：384
  • 距离度量：Cosine，L2
  • 价格：由 Hugging Face 收费
  • TiDB Cloud 托管：❌
  • Bring Your Own Key（BYOK，由用户自行提供 API 密钥）：✅
  • 项目主页：https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2

示例：

SET @@GLOBAL.TIDB_EXP_EMBED_HUGGINGFACE_API_KEY = 'your-huggingface-api-key-here';

CREATE TABLE sample (
  `id`        INT,
  `content`   TEXT,
  `embedding` VECTOR(384) GENERATED ALWAYS AS (EMBED_TEXT(
                "huggingface/sentence-transformers/all-MiniLM-L6-v2",
                `content`
              )) STORED
);


INSERT INTO sample
    (`id`, `content`)
VALUES
    (1, "Java: Object-oriented language for cross-platform development."),
    (2, "Java coffee: Bold Indonesian beans with low acidity."),
    (3, "Java island: Densely populated, home to Jakarta."),
    (4, "Java's syntax is used in Android apps."),
    (5, "Dark roast Java beans enhance espresso blends.");


SELECT `id`, `content` FROM sample
ORDER BY
  VEC_EMBED_COSINE_DISTANCE(
    embedding,
    "How to start learning Java programming?"
  )
LIMIT 2;

all-mpnet-base-v2

SET @@GLOBAL.TIDB_EXP_EMBED_HUGGINGFACE_API_KEY = 'your-huggingface-api-key-here';

CREATE TABLE sample (
  `id`        INT,
  `content`   TEXT,
  `embedding` VECTOR(768) GENERATED ALWAYS AS (EMBED_TEXT(
                "huggingface/sentence-transformers/all-mpnet-base-v2",
                `content`
              )) STORED
);


INSERT INTO sample
    (`id`, `content`)
VALUES
    (1, "Java: Object-oriented language for cross-platform development."),
    (2, "Java coffee: Bold Indonesian beans with low acidity."),
    (3, "Java island: Densely populated, home to Jakarta."),
    (4, "Java's syntax is used in Android apps."),
    (5, "Dark roast Java beans enhance espresso blends.");


SELECT `id`, `content` FROM sample
ORDER BY
  VEC_EMBED_COSINE_DISTANCE(
    embedding,
    "How to start learning Java programming?"
  )
LIMIT 2;

Qwen3-Embedding-0.6B

注意

Hugging Face Inference API 在该模型上可能不稳定。

  • 名称：huggingface/Qwen/Qwen3-Embedding-0.6B
  • 维度：1024
  • 距离度量：Cosine，L2
  • 最大输入文本 tokens：512
  • 价格：由 Hugging Face 收费
  • TiDB Cloud 托管：❌
  • Bring Your Own Key（BYOK，由用户自行提供 API 密钥）：✅
  • 项目主页：https://huggingface.co/Qwen/Qwen3-Embedding-0.6B
SET @@GLOBAL.TIDB_EXP_EMBED_HUGGINGFACE_API_KEY = 'your-huggingface-api-key-here';

CREATE TABLE sample (
  `id`        INT,
  `content`   TEXT,
  `embedding` VECTOR(1024) GENERATED ALWAYS AS (EMBED_TEXT(
                "huggingface/Qwen/Qwen3-Embedding-0.6B",
                `content`
              )) STORED
);


INSERT INTO sample
    (`id`, `content`)
VALUES
    (1, "Java: Object-oriented language for cross-platform development."),
    (2, "Java coffee: Bold Indonesian beans with low acidity."),
    (3, "Java island: Densely populated, home to Jakarta."),
    (4, "Java's syntax is used in Android apps."),
    (5, "Dark roast Java beans enhance espresso blends.");


SELECT `id`, `content` FROM sample
ORDER BY
  VEC_EMBED_COSINE_DISTANCE(
    embedding,
    "How to start learning Java programming?"
  )
LIMIT 2;

Python 使用示例

本示例展示如何创建向量表、插入文档，并使用 Hugging Face 向量嵌入模型进行相似度搜索。

步骤 1：连接数据库

from pytidb import TiDBClient

tidb_client = TiDBClient.connect(
    host="{gateway-region}.prod.aws.tidbcloud.com",
    port=4000,
    username="{prefix}.root",
    password="{password}",
    database="{database}",
    ensure_db=True,
)

步骤 2：配置 API 密钥

如果你使用私有模型或需要更高的速率限制，可以配置你的 Hugging Face API token。你可以在 Hugging Face Token Settings 页面创建 token：

通过 TiDB Client 为 Hugging Face 模型配置 API token：

tidb_client.configure_embedding_provider(
    provider="huggingface",
    api_key="{your-huggingface-token}",
)

步骤 3：创建向量表

创建一个包含向量字段的表，使用 Hugging Face 模型生成向量嵌入：

from pytidb.schema import TableModel, Field
from pytidb.embeddings import EmbeddingFunction
from pytidb.datatype import TEXT

class Document(TableModel):
    __tablename__ = "sample_documents"
    id: int = Field(primary_key=True)
    content: str = Field(sa_type=TEXT)
    embedding: list[float] = EmbeddingFunction(
        model_name="huggingface/sentence-transformers/all-MiniLM-L6-v2"
    ).VectorField(source_field="content")

table = tidb_client.create_table(schema=Document, if_exists="overwrite")

提示：

向量维度取决于你选择的模型。例如，huggingface/sentence-transformers/all-MiniLM-L6-v2 生成 384 维向量，而 huggingface/sentence-transformers/all-mpnet-base-v2 生成 768 维向量。

步骤 4：向表中插入数据

使用 table.insert() 或 table.bulk_insert() API 添加数据：

documents = [
    Document(id=1, content="Machine learning algorithms can identify patterns in data."),
    Document(id=2, content="Deep learning uses neural networks with multiple layers."),
    Document(id=3, content="Natural language processing helps computers understand text."),
    Document(id=4, content="Computer vision enables machines to interpret images."),
    Document(id=5, content="Reinforcement learning learns through trial and error."),
]
table.bulk_insert(documents)

步骤 5：搜索相似文档

使用 table.search() API 进行向量搜索：

results = table.search("How do neural networks work?") \
    .limit(3) \
    .to_list()

for doc in results:
    print(f"ID: {doc.id}, Content: {doc.content}")

另请参阅