o Zh'@sdZddlZddlmZddlmZmZmZmZm Z m Z ddl Z ddl Z ddlZddlmZddlmZddlmZmZmZdgZGd ddeeZGd d d ZdS) z-written under MIT Licence, Michael Feil 2023.N)ThreadPoolExecutor)AnyCallableDictListOptionalTuple) Embeddings)get_from_dict_or_env) BaseModel ConfigDictmodel_validatorInfinityEmbeddingsc@seZdZUdZeed< dZeed< dZeed< e ddZ e d d e d e d efd dZdeed eeefddZdeed eeefddZded eefddZded eefddZdS)raBSelf-hosted embedding models for `infinity` package. See https://github.com/michaelfeil/infinity This also works for text-embeddings-inference and other self-hosted openai-compatible servers. Infinity is a package to interact with Embedding Models on https://github.com/michaelfeil/infinity Example: .. code-block:: python from langchain_community.embeddings import InfinityEmbeddings InfinityEmbeddings( model="BAAI/bge-small", infinity_api_url="http://localhost:7997", ) modelzhttp://localhost:7997infinity_api_urlNclientZforbid)extrabefore)modevaluesreturncCs&t|dd|d<t|dd|d<|S)z?Validate that api key and python package exists in environment.rZINFINITY_API_URL)hostr)r &TinyAsyncOpenAIInfinityEmbeddingClient)clsrr^/var/www/html/lang_env/lib/python3.10/site-packages/langchain_community/embeddings/infinity.pyvalidate_environment3s z'InfinityEmbeddings.validate_environmenttextscCs|jj|j|d}|S)zCall out to Infinity's embedding endpoint. Args: texts: The list of texts to embed. Returns: List of embeddings, one for each text. rr)rembedrselfr embeddingsrrrembed_documentsAs  z"InfinityEmbeddings.embed_documentscs|jj|j|dIdH}|S)zAsync call out to Infinity's embedding endpoint. Args: texts: The list of texts to embed. Returns: List of embeddings, one for each text. rN)raembedrr rrraembed_documentsPs   z#InfinityEmbeddings.aembed_documentstextcCs||gdS)zCall out to Infinity's embedding endpoint. Args: text: The text to embed. Returns: Embeddings for the text. r)r#)r!r&rrr embed_query_s zInfinityEmbeddings.embed_querycs||gIdH}|dS)zAsync call out to Infinity's embedding endpoint. Args: text: The text to embed. Returns: Embeddings for the text. Nr)r%)r!r&r"rrr aembed_queryjs zInfinityEmbeddings.aembed_query)__name__ __module__ __qualname____doc__str__annotations__rrrr Z model_configr classmethodrrrfloatr#r%r'r(rrrrrs$     c @sPeZdZdZ  d dedeejddfddZe e fd e ed e de e ee ffd d Zd e ede e efd dZe de e ede efddZded e edeeeffddZdede ede e efddZded e ede e efddZdejdeeefde e efddZded e ede e efddZdS)!ra Helper tool to embed Infinity. It is not a part of Langchain's stable API, direct use discouraged. Example: .. code-block:: python mini_client = TinyAsyncInfinityEmbeddingClient( ) embeds = mini_client.embed( model="BAAI/bge-small", text=["doc1", "doc2"] ) # or embeds = await mini_client.aembed( model="BAAI/bge-small", text=["doc1", "doc2"] ) http://localhost:7797/v1Nr aiosessionrcCs6||_||_|jdust|jdkrtdd|_dS)Nz( param `host` must be set to a valid url)rr2len ValueError _batch_size)r!rr2rrr__init__s  z/TinyAsyncOpenAIInfinityEmbeddingClient.__init__rsortercsRtdkr ddfStfddDfddD}|fddfS)aSort texts in ascending order, and delivers a lambda expr, which can sort a same length list https://github.com/UKPLab/sentence-transformers/blob/ c5f93f70eca933c78695c5bc686ceda59651ae3b/sentence_transformers/SentenceTransformer.py#L156 Args: texts (List[str]): _description_ sorter (Callable, optional): _description_. Defaults to len. Returns: Tuple[List[str], Callable]: _description_ Example: ``` texts = ["one","three","four"] perm_texts, undo = self._permute(texts) texts == undo(perm_texts) ``` cSs|S)Nr)trrrszATinyAsyncOpenAIInfinityEmbeddingClient._permute..csg|]}| qSrr).0Zsen)r9rr szCTinyAsyncOpenAIInfinityEmbeddingClient._permute..cg|]}|qSrrr=idx)rrrr>csfddtDS)Ncr?rrr@Zunsorted_embeddingsrrr>szUTinyAsyncOpenAIInfinityEmbeddingClient._permute....)npargsortrC)length_sorted_idxrCrr<s )r5rDrE)rr9Z texts_sortedr)rFr9rr_permutes  z/TinyAsyncOpenAIInfinityEmbeddingClient._permutecCsJt|dkr |gSg}tdt||jD]}|||||jq|S)aX splits Lists of text parts into batches of size max `self._batch_size` When encoding vector database, Args: texts (List[str]): List of sentences self._batch_size (int, optional): max batch size of one request. Returns: List[List[str]]: Batches of List of sentences r:r)r5ranger7append)r!rZbatches start_indexrrr_batchs z-TinyAsyncOpenAIInfinityEmbeddingClient._batchbatch_of_textscCs@t|dkrt|ddkr|dSg}|D]}||q|S)Nr:r)r5extend)rLrZsublistrrr_unbatchs  z/TinyAsyncOpenAIInfinityEmbeddingClient._unbatchrcCs"t|jdddit||ddS)zBuild the kwargs for the Post request, used by sync Args: model (str): _description_ texts (List[str]): _description_ Returns: Dict[str, Collection[str]]: _description_ z /embeddingsz content-typezapplication/json)inputr)urlheadersjson)dictr)r!rrrrr_kwargs_post_requests z;TinyAsyncOpenAIInfinityEmbeddingClient._kwargs_post_request batch_textscCsRtjdi|j||d}|jdkrtd|jd|jdd|dDS) Nr5Infinity returned an unexpected response with status : cSg|]}|dqS embeddingrr=errrr>rBzNTinyAsyncOpenAIInfinityEmbeddingClient._sync_request_embed..datar)requestspostrT status_code Exceptionr&rR)r!rrUresponserrr_sync_request_embeds  z:TinyAsyncOpenAIInfinityEmbeddingClient._sync_request_embedc Cs||\}}||}|j|gt||f}t|dkr$tt|}ntd}t|j|}Wdn1s:wY||} || } | S)zcall the embedding of model Args: model (str): to embedding model texts (List[str]): List of sentences to embed. Returns: List[List[float]]: List of vectors for each sentence r: N)rGrKrdr5listmaprrN) r!rr perm_textsunpermute_funcperm_texts_batchedZmap_argsembeddings_batch_permpembeddings_permr"rrrrs     z,TinyAsyncOpenAIInfinityEmbeddingClient.embedsessionkwargsc s|jdi|4IdH,}|jdkrtd|jd|j|IdHd}dd|DWdIdHS1IdHs?wYdS)NrVrWrXr^cSrYrZrr\rrrr>%rBzITinyAsyncOpenAIInfinityEmbeddingClient._async_request..r)r`statusrbr&rR)r!rnrorcr[rrr_async_requests  0z5TinyAsyncOpenAIInfinityEmbeddingClient._async_requestc s|\}}|}tjdtjddd4IdHtjfdd|DIdH}WdIdHn 1IdHs>wY|}||}|S)zcall the embedding of model, async method Args: model (str): to embedding model texts (List[str]): List of sentences to embed. Returns: List[List[float]]: List of vectors for each sentence Tre)limit) trust_envZ connectorNcs$g|]}jj|ddqS)r)rnro)rqrT)r=r;rr!rnrrr>9s zATinyAsyncOpenAIInfinityEmbeddingClient.aembed..)rGrKaiohttp ClientSessionZ TCPConnectorasynciogatherrN) r!rrrhrirjrkrmr"rrtrr$'s    ( z-TinyAsyncOpenAIInfinityEmbeddingClient.aembed)r1N)r)r*r+r,r-rrurvr8 staticmethodr5rrrrGrKrrNrrTr0rdrrqr$rrrrrwsP   ! "   "    & r)r,rwconcurrent.futuresrtypingrrrrrrrunumpyrDr_Zlangchain_core.embeddingsr Zlangchain_core.utilsr Zpydanticr r r __all__rrrrrrs    f