i"PJSSKJr SSKrSSKrSSKJrJr SSKrSSK J r \R"\ 5r \(aSSKrSSKrSS SjjrSS SjjrS S Sjjrg) ) annotationsN) TYPE_CHECKINGLiteral)Tensorc SSKn UbUb [S5eUcUc [S5eUcUS;aHU (a U R"URS5nOtU R"URSS5nOSUS:XaMU (a#U R "URSS -5nO#U R "URSS -S5nURU5 Sn Un UR[R[R4;a6U(a"US :waUn X-n O[RS 5 [UUUUS 9nOU(a[RS 5 [R"5nUR!X 5unnU Gb)[R""UVVs/sH3nUVs/sH"nUR%UR'55PM$ snPM5 snn5nUS:Xa)[R("USS9R+[,5nOUR+[,5n[R."SU U5n[R0"U*5SS2SU24nU[R2"[5U55SS2S4U4nU[R2"[5U55SS2S4U4n[R"5U- n[7[5U55VVVs/sH@n[9UUUU5VVs/sHunn[-U5[;U5S.PM snnPMB snnnU4nU (a/UQUP7nU$s snfs snnfs snnfs snnnf)aF Performs semantic search using the FAISS library. Rescoring will be performed if: 1. `rescore` is True 2. The query embeddings are not quantized 3. The corpus is quantized, i.e. the corpus precision is not float32 Only if these conditions are true, will we search for `top_k * rescore_multiplier` samples and then rescore to only keep `top_k`. Args: query_embeddings: Embeddings of the query sentences. Ideally not quantized to allow for rescoring. corpus_embeddings: Embeddings of the corpus sentences. Either `corpus_embeddings` or `corpus_index` should be used, not both. The embeddings can be quantized to "int8" or "binary" for more efficient search. corpus_index: FAISS index for the corpus sentences. Either `corpus_embeddings` or `corpus_index` should be used, not both. corpus_precision: Precision of the corpus embeddings. The options are "float32", "int8", or "binary". Default is "float32". top_k: Number of top results to retrieve. Default is 10. ranges: Ranges for quantization of embeddings. This is only used for int8 quantization, where the ranges refers to the minimum and maximum values for each dimension. So, it's a 2D array with shape (2, embedding_dim). Default is None, which means that the ranges will be calculated from the calibration embeddings. calibration_embeddings: Embeddings used for calibration during quantization. This is only used for int8 quantization, where the calibration embeddings can be used to compute ranges, i.e. the minimum and maximum values for each dimension. Default is None, which means that the ranges will be calculated from the query embeddings. This is not recommended. rescore: Whether to perform rescoring. Note that rescoring still will only be used if the query embeddings are not quantized and the corpus is quantized, i.e. the corpus precision is not "float32". Default is True. rescore_multiplier: Oversampling factor for rescoring. The code will now search `top_k * rescore_multiplier` samples and then rescore to only keep `top_k`. Default is 2. exact: Whether to use exact search or approximate search. Default is True. output_index: Whether to output the FAISS index used for the search. Default is False. Returns: A tuple containing a list of search results and the time taken for the search. If `output_index` is True, the tuple will also contain the FAISS index used for the search. Raises: ValueError: If both `corpus_embeddings` and `corpus_index` are provided or if neither is provided. The list of search results is in the format: [[{"corpus_id": int, "score": float}, ...], ...] The time taken for the search is a float value. rN@Only corpus_embeddings or corpus_index should be used, not both.8Either corpus_embeddings or corpus_index should be used.)float32uint8ubinaryr zRescoring is enabled but the corpus is not quantized. Either pass `rescore=False` or quantize the corpus embeddings with `quantize_embeddings(embeddings, precision="...") `and pass `corpus_precision="..."` to `semantic_search_faiss`. precisionrangescalibration_embeddingsRescoring is enabled but the query embeddings are quantized. Either pass `rescore=False` or don't quantize the query embeddings.axis ij,ikj->ik corpus_idscore)faiss ValueError IndexFlatIPshape IndexHNSWFlatIndexBinaryFlatIndexBinaryHNSWadddtypenpr int8loggerwarningquantize_embeddingstimesearcharray reconstructitem unpackbitsastypeinteinsumargsortarangelenrangezipfloat)query_embeddingscorpus_embeddings corpus_indexcorpus_precisiontop_krrrescorerescore_multiplierexact output_indexrrescore_embeddingskstart_tscoresindices query_indicesidxtop_k_embeddingsrescored_scoresrescored_indicesdelta_tquery_idrneighboroutputss r/home/dmtnaga/Documents/work/airagagent/rag_env/lib/python3.13/site-packages/sentence_transformers/quantization.pysemantic_search_faissrQsBT$)A[\\ \%9STT 3 3$001B1H1H1KL $223D3J3J13MrR  *$445F5L5LQ5ORS5ST $445F5L5LQ5ORS5SUWX *+  Abhh%88 9,%5"'T / &#9    O iikG"))*:>OFG%88bi jbiQ^m Lmsl&&sxxz2m Lbi j  y (!}}-=BGNNsS /66s; ))L2DFVW::&676E6 B"))C(8$9:1d7CEUUV 3/?+@!A!T'!JL\!\]iikG#G"#&6"78  9(+6(+;WX=N'O 'OOE8"(meElC'O 9    G*G*\* NAM j,  s0" M ,)MM &M$M%MM Mc tSSKJn SSKJn UbUb [ S5eUcUc [ S5eUS;a [ S5eUcUS :XaU "UR S S S S 9nOVUS:XaU "UR S SSS 9nO:US:XaU "UR S SSS 9nOUS:XaU "UR S S-SSS 9nUR [R"[U55U5 Sn UnUR[R[R4;aJU(a6URU R:waUn X-nO[RS5 [!UUUUS9nOU(a[RS5 ["R""5nUR%XU S9nUR&nUR(nUR*S:a[R,"U5nUR*S:a[R,"U5nU Gb[R."UVs/sHnUR1U5PM sn5nUS;a2[R2"UR5[R5SS9nUR5[65n[R8"SU U5n[R:"U*5SS2SU24nU[R"[U55SS2S4U4nU[R"[U55SS2S4U4n["R""5U- n[=[U55VVVs/sH@n[?UUUU5VVs/sHunn[7U5[AU5S.PM snnPMB snnnU4nU (a/UQUP7nU$s snfs snnfs snnnf) aX Performs semantic search using the usearch library. Rescoring will be performed if: 1. `rescore` is True 2. The query embeddings are not quantized 3. The corpus is quantized, i.e. the corpus precision is not float32 Only if these conditions are true, will we search for `top_k * rescore_multiplier` samples and then rescore to only keep `top_k`. Args: query_embeddings: Embeddings of the query sentences. Ideally not quantized to allow for rescoring. corpus_embeddings: Embeddings of the corpus sentences. Either `corpus_embeddings` or `corpus_index` should be used, not both. The embeddings can be quantized to "int8" or "binary" for more efficient search. corpus_index: usearch index for the corpus sentences. Either `corpus_embeddings` or `corpus_index` should be used, not both. corpus_precision: Precision of the corpus embeddings. The options are "float32", "int8", "ubinary" or "binary". Default is "float32". top_k: Number of top results to retrieve. Default is 10. ranges: Ranges for quantization of embeddings. This is only used for int8 quantization, where the ranges refers to the minimum and maximum values for each dimension. So, it's a 2D array with shape (2, embedding_dim). Default is None, which means that the ranges will be calculated from the calibration embeddings. calibration_embeddings: Embeddings used for calibration during quantization. This is only used for int8 quantization, where the calibration embeddings can be used to compute ranges, i.e. the minimum and maximum values for each dimension. Default is None, which means that the ranges will be calculated from the query embeddings. This is not recommended. rescore: Whether to perform rescoring. Note that rescoring still will only be used if the query embeddings are not quantized and the corpus is quantized, i.e. the corpus precision is not "float32". Default is True. rescore_multiplier: Oversampling factor for rescoring. The code will now search `top_k * rescore_multiplier` samples and then rescore to only keep `top_k`. Default is 2. exact: Whether to use exact search or approximate search. Default is True. output_index: Whether to output the usearch index used for the search. Default is False. Returns: A tuple containing a list of search results and the time taken for the search. If `output_index` is True, the tuple will also contain the usearch index used for the search. Raises: ValueError: If both `corpus_embeddings` and `corpus_index` are provided or if neither is provided. The list of search results is in the format: [[{"corpus_id": int, "score": float}, ...], ...] The time taken for the search is a float value. r) ScalarKind)IndexNrr )r r&rbinaryzKcorpus_precision must be "float32", "int8", "ubinary", "binary" for usearchr r cosf32)ndimmetricr$r&ipi8rUhammingrrb1zRescoring is enabled but the corpus is not quantized. Either pass `rescore=False` or quantize the corpus embeddings with `quantize_embeddings(embeddings, precision="...") `and pass `corpus_precision="..."` to `semantic_search_usearch`.rr)countr@)rrUrrrr)!usearch.compiledrS usearch.indexrTrrr#r%r4r5r$r r&F32r'r(r)r*r+ distanceskeysrX atleast_2dr,getr/r0r1r2r3r6r7r8)r9r:r;r<r=rrr>r?r@rArSrTrBrCrDmatchesrErFrGrIrJrKrLrMrrNrOs rPsemantic_search_usearchrhsT,#$)A[\\ \%9STTGGfgg y ( &,,Q/L  ' &,,Q/L  ) &,,Q/ L  * &,,Q/!3 L 3'8#9:%1BFF+A$JBFFSipqLr#st##A&,NN$YK/NsS]N__iqturAEFrFjmLNjOO%;'cc BFF:A$>zXY@Z#[\11q!t ,3  (E199"((C C & (E1C7??H HH JN+33J4D4DQ4GLsRZZ[][b[bccI{{:>*22:3C3CA3FKK z).?@ AAGOs&'L ) NNr NNTr_TF)r9 np.ndarrayr:np.ndarray | Noner;zfaiss.Index | Noner<z&Literal['float32', 'uint8', 'ubinary']r=r1rrrrr>boolr?r1r@rrArreturnz=tuple[list[list[dict[str, int | float]]], float, faiss.Index])r9rr:rr;zusearch.index.Index | Noner<z$Literal['float32', 'int8', 'binary']r=r1rrrrr>rr?r1r@rrArrzEtuple[list[list[dict[str, int | float]]], float, usearch.index.Index])NN) rzzTensor | np.ndarrayrz8Literal['float32', 'int8', 'uint8', 'binary', 'ubinary']rrrrrr) __future__rloggingr*typingrrrqr%torchr getLogger__name__r'rusearchrQrhr)rPrs" )   8 $ ,0'+?H $04d d(d%d= d  d  d.ddd ddCdR,0/3=F $04w w(w-w; w  w  w.www wwKwz!%04 GB#GBGGB GB. GB  GBr