Ig ddlmZmZmZmZmZmZmZmZddl Z ddl m Z ddl mZddlmZddlmZmZddlmZmZdd d d d d didddd d d d d diddgZGddeeeZGddeZy))AnyDictIterableListOptionalSequenceTypeUnionN) OnnxProvider)OnnxOutputContext)define_cache_dir)SparseEmbeddingSparseTextEmbeddingBase) OnnxTextModelTextEmbeddingWorkerzprithivida/Splade_PP_en_v1i:wz9Independent Implementation of SPLADE++ Model for English.z apache-2.0g/$?hfzQdrant/SPLADE_PP_en_v1z model.onnx)model vocab_size descriptionlicense size_in_GBsources model_filezprithvida/Splade_PP_en_v1ceZdZdedeefdZedee e e ffdZ dde de e de ede eed ed e eed ed e effd ZddZ ddee ee fdede edeefdZedeefdZxZS)SpladePPoutputreturnc#xK|j tdtjdtj|j dz}|tj |jdz}tj|d}|D])}|jd}||}t||+yw)Nz/ e.g. BAAI/bge-base-en. Nr)local_files_only)super__init__r=r@r?r>rA_get_model_descriptionmodel_descriptionr r;download_model_local_files_only _model_dirload_onnx_model) r-r:r;r<r=r>r?r@rAkwargs __class__s r2rEzSpladePP.__init__Bs@ YB6B""%   &DN __ (!__Q/DN!DN!%!rA)_load_onnx_modelrJrGr<r=r>rA)r-s r2rKzSpladePP.load_onnx_model|sF oo--l;LLnnnn  r9 documents batch_sizeparallelc +K|jd|jt|j||||j|j |j d|Ed{y7w)a Encode a list of documents into list of embeddings. We use mean pooling with attention so that the model can handle variable-length inputs. Args: documents: Iterator of documents or single document to embed batch_size: Batch size for encoding -- higher values will use more memory, but be faster parallel: If > 1, data-parallel encoding will be used, recommended for offline encoding of large datasets. If 0, use all available cores. If None, don't use data-parallel processing, use default onnxruntime threading instead. Returns: List of embeddings, one per document )r:r;rQrRrSr=r>r?N)_embed_documentsr:strr;r=r>r?)r-rQrRrSrLs r2embedzSpladePP.embedsb,)4((  $..)!nn    sAA$A"A$ctS)N)SpladePPEmbeddingWorkerr6s r2_get_worker_classzSpladePP._get_worker_classs&&r9)NNNFNFN)rN)N)__name__ __module__ __qualname__r rrr3 classmethodrrrWrr8rintrr boolrErKr rXr rr[ __classcell__)rMs@r2rr'sBB0ABhF_B"'d4S>&:''$(!%6:*.#'8#8#C=8## 8# H\23 8#  8#T#Y'8#8#C=8#t "&  hsm+,    3-  / "  D'$':";''r9rc eZdZdededefdZy)rZr:r;rc td||dd|S)Nr)r:r;r<rU)r)r-r:r;rLs r2init_embeddingz&SpladePPEmbeddingWorker.init_embeddings' !    r9N)r]r^r_rWrrfrUr9r2rZrZs   8 r9rZ)typingrrrrrrr r numpyr&fastembed.commonr fastembed.common.onnx_modelr fastembed.common.utilsr &fastembed.sparse.sparse_embedding_baserrfastembed.text.onnx_text_modelrrr5rrZrUr9r2rnsMMM)93N.R * # -R * # 4C'& o(FC'L 1 r9