i(rdZddlmZddlZddlmZddlmZddlm Z e e Z GddZ dS) uEmbedding client using an OpenAI-compatible embeddings API. 文本向量化客户端封装模块。 通过 OpenAI 兼容的 Embeddings API 将文本转换为稠密向量， 支持批量编码和单条查询编码，默认使用 bge-m3 模型（1024 维）。 ) annotationsN) AsyncOpenAI)settings) get_loggercJeZdZdZddddddd Zdd ZddddZddddZdS)EmbeddingClientuAsync wrapper for computing text embeddings via an OpenAI-compatible API. 异步文本向量化客户端，支持批量和单条文本的向量计算。 N)base_urlapi_keymodel dimensionsr str | Noner r r int | NonereturnNonec *|p tj|_|p tj|_t |p tj|p tjtj ddtj dtj dd|_ dS)Ng$@g@)connectF) trust_envtimeout)r r r http_client) rembedding_model_modelembedding_dimensions _dimensionsrembedding_base_urlembedding_api_keyhttpxTimeout AsyncClient_client)selfr r r r s =D:\work\zm-rag\backend\app\infrastructure\embedding_client.py__init__zEmbeddingClient.__init__s7x7 %F)F#<!<9x9M$444) dC888     cHK|jd{VdS)N)rclose)r s r!r%zEmbeddingClient.close0s2l  """""""""""r#r texts list[str]list[list[float]]c K|sgS |jj|p|j||jd{V}nR#t $rE}t d|p|jt|t|d}~wwxYwd|j D}t|D]\}}t||jkr|t d|jt||t||jkr|d|j||<x|dg|jt|z zz||<|S) aEmbed a batch of texts and return a list of vectors. Each vector has *self._dimensions* dimensions (default 1024 for ``bge-m3``). The OpenAI embeddings API accepts a list of strings so a single round-trip is made for the whole batch. )r inputr Nembedding_api_error)r text_counterrorcg|] }|j S) embedding).0items r! z/EmbeddingClient.embed_texts..Vs<<<< list[float]cRK||g|d{V}|dS)zEmbed a single query text and return the vector. This is a convenience wrapper around :meth:`embed_texts` for the common single-query use case. r&Nr)rG)r rHr rDs r! embed_queryzEmbeddingClient.embed_queryks>(($u(========qzr#) r r r r r r r rrr)rr)r'r(r r rr))rHr>r r rrI)__name__ __module__ __qualname____doc__r"r%rGrKr0r#r!rrs $" !%       .####! 666666x!        r#r) rO __future__rropenair app.configrapp.utils.loggerrrLr<rr0r#r!rTs#""""" '''''' H  ddddddddddr#