类class agentscope.embedding

GeminiEmbeddingModel

class GeminiEmbeddingModel(EmbeddingModelBase[str | TextBlock | DataBlock])

源码Source agentscope/embedding/_gemini/_model.py:67–487

Gemini 向量化模型，与模型服务通信并返回生成或向量化结果。位于 agentscope.embedding 包，继承自 EmbeddingModelBase[str | TextBlock | DataBlock]。

Routes to the text-only or multimodal Gemini API based on the model name.

Text mode (gemini-embedding-001): uses the base class's simple batch splitting + concurrent retry. The Gemini API accepts a list of strings and returns individual embeddings.
Multimodal mode (gemini-embedding-2): overrides __call__ with content-aware batching (respecting per-model limits on images, videos, audios, PDFs). Each input is wrapped in a Content object so the API returns separate embeddings.

Key API differences from other providers:

Dimensions are controlled via output_dimensionality in the config parameter (not a top-level dimensions field).
gemini-embedding-001 supports task_type in config; gemini-embedding-2 uses prompt prefixes instead.

构造器摘要Constructor Summary

构造器Constructor	说明Description
`def __init__(credential:` `CredentialBase, model: str, dimensions: int \| None, parameters: 'GeminiEmbeddingModel.Paramete…`	初始化 `GeminiEmbeddingModel` 实例。 Initialize the Gemini embedding model.

方法摘要Method Summary

方法Method	说明Description
`async def __call__(inputs: list[str \|` `TextBlock \| DataBlock], **kwargs: Any) -> EmbeddingResponse`async	以可调用对象方式执行。 Embed inputs with batching and retry.

构造器详情Constructor Detail

GeminiEmbeddingModel

def __init__(credential: CredentialBase, model: str, dimensions: int | None, parameters: 'GeminiEmbeddingModel.Parameters | None' = None, embedding_cache: EmbeddingCacheBase | None = None, context_size: int = 8192, max_retries: int = 3, retry_delay: float = 1.0) -> None

初始化 GeminiEmbeddingModel 实例。

Initialize the Gemini embedding model.

参数Parameters

名称Name	类型Type	说明Description
`credential`	CredentialBase	凭证对象（类型 `CredentialBase`）。A :class:`~agentscope.credential.GeminiCredential` instance providing the API key.
`model`	str	模型实例（类型 `str`）。The embedding model name (e.g. `"gemini-embedding-001"` or `"gemini-embedding-2"`).
`dimensions`	int \| None	向量维度（类型 `int \| None`）。The output embedding vector dimensions. Required at the contract level — see :class:`EmbeddingModelBase` for the rationale. `None` is accepted only for backward compatibility with legacy configs that persisted `dimensions` inside `parameters`.
`parameters`默认default `None`	'GeminiEmbeddingModel.Parameters \| None'	参数（类型 `'GeminiEmbeddingModel.Parameters \| None'`）。Provider-specific non-dimensional text embedding options, such as `task_type`. Per-call options take precedence.
`embedding_cache`默认default `None`	EmbeddingCacheBase \| None	向量化缓存（类型 `EmbeddingCacheBase \| None`）。Optional embedding cache.
`context_size`默认default `8192`	int	上下文大小（类型 `int`）。Maximum input tokens. 2048 for `gemini-embedding-001`, 8192 for `gemini-embedding-2`.
`max_retries`默认default `3`	int	最大重试次数（类型 `int`）。Number of retries on transient failures.
`retry_delay`默认default `1.0`	float	重试延迟（类型 `float`）。Seconds between retry attempts.

源码Source agentscope/embedding/_gemini/_model.py:93–157

方法详情Method Detail

call

async def __call__(inputs: list[str | TextBlock | DataBlock], **kwargs: Any) -> EmbeddingResponse

async

以可调用对象方式执行。

For text models, delegates to the base class.

参数Parameters

名称Name	类型Type	说明Description
`inputs`	list[str \| TextBlock \| DataBlock]	输入消息（类型 `list[str \| TextBlock \| DataBlock]`）。The input data to embed. `TextBlock` items are unpacked to their `.text` field on entry, so the remainder of the pipeline only sees `str` and `DataBlock`.
`kwargs`	Any	关键字参数（类型 `Any`）。The kwargs.

返回Returns EmbeddingResponse

返回EmbeddingResponse。`EmbeddingResponse`: Merged response for all inputs.

源码Source agentscope/embedding/_gemini/_model.py:163–209

继承成员Inherited Members

继承自Inherited from EmbeddingModelBase

list_models, credential, model, dimensions, context_size, batch_size, max_retries, retry_delay, supports_multimodal