rhpddlmZddlZddlZddlZddlZddlmZmZddl Z ddl Z ddl m Z ddlm Z ddlmZej"eZerddlmZdd lmZdd lmZ d d d Zy)) annotationsN) TYPE_CHECKINGLiteral)trange)tqdm)is_datasets_availableDataset) CrossEncoder)SentenceTransformerc(ts tdt|dk(r tdddlm}|j }|r||vr|d}|r||vr|d}|s|st|dk7r td||rd nd }tjd |d |rA|dk7s | ||| d k7rtjd|d k7rtjdd }|dvrtd|d||} tjd| d t|jj|jj|j}t|}|A| | | || dzz|z}n|| z|z}|dkDr|rd}|r t!d|rt!d|di} t||}!t||}"|du}#|#s|"}tt"j%||"z}t'|D$%cic]\}$}%|%|$ }&}$}%|!j)}'tt"j%|!}!t'|!D$(cic]\}$}(|(|$ })}$}(t|!}*t+j,|*j/d}+|j0},|*t|'k7r|rt!d|*dt|'d|dkDr%t3j4|}-t!d|-ddd}.d}/|rt7j8|d !|j:j<xsd"}0t?j@|0d"jC|!zjEd#$jG}1t?j@|0d"jC|zjEd#$jG}2t6jHjC|d%|1d&}3t6jHjC|d'|2d&}4t6jHjK|3r3t3jL|3}/|rt!d(|3d)|/jNd*t6jHjK|4r3t3jL|4}.|rt!d+|4d)|.jNd*|.|/|rl|jQtS|tTrdn|,}5|.|jE||5|d d d ||-}.|/|jE|!|5|d d d ||-}/|jW|5n4|.|jE||d d d ||.}.|/|jE|!|d d d ||.}/|rt6jHjK3s&t3jX|3|/|rt!d/|3t6jHjK4s&t3jX|4|.|rt!d0|4|r=ddl-}6|6j]|j_}7 |6ja}8d |8_1d |8_2|6jg|7|81}7|7jk|.g}9g}:tmdt|/|d23D]E};|/|;|;|z}<|7jo|<|dz4\}=}>|9jq|=|:jq|>Gt+jrt3jt|9d5jw|,}=t+jrt3jt|:d5jw|,}>n?|jy|/|.jw|,}=t+jz|=||zd6\}=}>t}|*D?cgc]}?g}@}?t|'|"D]!\}(}A|)|(}B@|Bjq|&|A#@DCcgc] }Ct|C}D}Cg}"g}'t}|*D]B}$|"j@|$DEcgc]}E||E c}E|'j|!|$gD|$zD@DCcgc]}Ct+j|C|,7}@}C|/t}|*D$?cgc]}$t}D|$D]}?|$c}?}$}/|.t+j@j}F|j|/|Fjw|,}G~/~F~.|| | |tt'|>d8t|>9D]O\}$}H|!|$}(|HDIcgc]}I||I }J}I|jtt|(g|dzz|J|d :}K|K|=|$<Q|jtt|'|"|d :}G|sUt+jt}|*DLcgc]}Lt+j|>|L@|L c}L}Mtd; |=|M<|=j}N| | Qt+j|*Gj0|Gj<}Od}Pt}|*D].}Lt+jGP|PD|LzO|L<|P|D|Ldz z }P0| p|=| zOj|=jddjkD}Qtd; |=|Q<|Qjj}R|RrR|RNz d=| d><|N|Rz}N| s|=Oj|=jddjd| z zkD}Qtd; |=|Q<|Qjj}R|RrR|RNz d=| d?<|N|Rz}N|?|=|kD}Qtd; |=|Q<|Qjj}R|Rr R|RNz d=| d@<| ?|=| k}Qtd; |=|Q<|Qjj}R|Rr R|RNz d=| dA<t+jz|=|d6\}S}T|>|+|Tf}>|r|>dd|df}>Sdd|df}S| d k(r|>ddd| f}>Sddd| f}Sn| dBk(r|>jdSjjdz }U|Uj| C}U|UDVcgc]"}Vtjt}|V| 4$}W}V|>|+|Wf}>S|+|Wf}S|Sjdd D\}S}T|>|+|Tf}>t+jt}|*D$cgc]}$|>|$jD|$dc}$}>t+jt}|*D$cgc]}$S|$jD|$dc}$}S|r t!dE|d k(rGStd; k7}Xt+j|>}Yt+j|>}Z|>|X}>|S|X}Sd}Pt}|*D]c}Lt+j|LjD|L| YP|P|D|Lz@|Lj| djZ|P|P|D|Lz|P|D|Lz }PeYX}YZ|X}@|g|gdFgi}[t|Y|@|>D]K\}\}]}^[|jq|!|\|[|jq||]|[dFjq||^MGj| djXSz }_nW|dGk(rStd; k7}X|g|gdHgi}[t}|*D]}B@|BD]D}][|jq|!B|[|jq||]|[dHjqdFt|>BS|BD]W\}^}`|`td; k(r[|jq|!B|[|jq|^|[dHjqdYSX}SGj| dj|X|Sz }_nI|dIvr7Std; k7jdJ}X|S|X}S|>|X}>|t'|XD$acgc] \}$}a|as |'|$c}a}$|t'XD$acgc] \}$}a|as |"|$c}a}$it'|>jdKD;bccic]\};}bdL|;|bDccgc]}c||c c}cc}c}b};}[|dMk(rD$dea7cgc]D\}$\}d}e|djr,|"|$gtdeDa7cgc] \}a}7|as ||7c}7}azFc}7}a}e}d}$dPXDdcgc]&}d|djsdgdgtdzz(c}di}[SX}SGj| dj|X|Sz }_t[dk(r tdQ|j[}f|rdR}gdS}ht!|gjdTdUdVdWt!|gjdX|htG|htSd"dYt*j4fdZt*jfd[t*jfd\d]fd^d_fd`dafdbdcfdddeffD]T\}i}jt!gj|ijh|jG|h|jS|h|j_Vd>| fd?| fd@|fdA| ffD]4\}k}l|k| vs | kdf}m| |kdg}nt!dh|mdidj|ndkdl|kdmldn 6| t|ztSz }o|odkDrdog}p|dkDrpjqdp| pjqd>| pjqd?|pjqd@dqjCpdd}qt|pdkDr qdrpdzz }qo| t|zz }rt!ds|odt|rdkduqdvtpdkDrdwnd"dx fScc}%}$wcc}(}$w#th$rY wxYwcc}?wcc}Cwcc}Ewcc}Cwcc}?}$wcc}Iwcc}Lwcc}Vwcc}$wcc}$wcc}a}$wcc}a}$wcc}cwcc}c}b};wcc}d}$wcc}7}awcc}7}a}e}d}$wcc}dw)yu> Add hard negatives to a dataset of (anchor, positive) pairs to create (anchor, positive, negative) triplets or (anchor, positive, negative_1, ..., negative_n) tuples. Hard negative mining is a technique to improve the quality of a dataset by adding hard negatives, which are texts that may appear similar to the anchor, but are not. Using hard negatives can improve the performance of models trained on the dataset. This function uses a SentenceTransformer model to embed the sentences in the dataset, and then finds the closest matches to each anchor sentence in the dataset. It then samples negatives from the closest matches, optionally using a CrossEncoder model to rescore the candidates. Supports prompt formatting for models that expect specific instruction-style input. You can influence the candidate negative selection in various ways: - **range_min**: Minimum rank of the closest matches to consider as negatives: useful to skip the most similar texts to avoid marking texts as negative that are actually positives. - **range_max**: Maximum rank of the closest matches to consider as negatives: useful to limit the number of candidates to sample negatives from. A lower value makes processing faster, but may result in less candidate negatives that satisfy the margin or max_score conditions. - **max_score**: Maximum score to consider as a negative: useful to skip candidates that are too similar to the anchor. - **min_score**: Minimum score to consider as a negative: useful to skip candidates that are too dissimilar to the anchor. - **absolute_margin**: Absolute margin for hard negative mining: useful to skip candidate negatives whose similarity to the anchor is within a certain margin of the positive pair. A value of 0 can be used to enforce that the negative is always further away from the anchor than the positive. - **relative_margin**: Relative margin for hard negative mining: useful to skip candidate negatives whose similarity to the anchor is within a certain margin of the positive pair. A value of 0.05 means that the negative is at most 95% as similar to the anchor as the positive. - **sampling_strategy**: Sampling strategy for negatives: "top" or "random". "top" will always sample the top n candidates as negatives, while "random" will sample n negatives randomly from the candidates that satisfy the margin or max_score conditions. .. tip:: The excellent `NV-Retriever paper `_ is a great resource for understanding the details of hard negative mining and how to use it effectively. Notably, it reaches the strongest performance using these settings:: dataset = mine_hard_negatives( dataset=dataset, model=model, relative_margin=0.05, # 0.05 means that the negative is at most 95% as similar to the anchor as the positive num_negatives=num_negatives, # 10 or less is recommended sampling_strategy="top", # "top" means that we sample the top candidates as negatives batch_size=batch_size, # Adjust as needed use_faiss=True, # Optional: Use faiss/faiss-gpu for faster similarity search ) This corresponds with the `TopK-PercPos (95%)` mining method. Example: >>> from sentence_transformers.util import mine_hard_negatives >>> from sentence_transformers import SentenceTransformer >>> from datasets import load_dataset >>> # Load a Sentence Transformer model >>> model = SentenceTransformer("all-MiniLM-L6-v2") >>> >>> # Load a dataset to mine hard negatives from >>> dataset = load_dataset("sentence-transformers/natural-questions", split="train") >>> dataset Dataset({ features: ['query', 'answer'], num_rows: 100231 }) >>> dataset = mine_hard_negatives( ... dataset=dataset, ... model=model, ... range_min=10, ... range_max=50, ... max_score=0.8, ... relative_margin=0.05, ... num_negatives=5, ... sampling_strategy="random", ... batch_size=128, ... use_faiss=True, ... ) Batches: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████| 588/588 [00:32<00:00, 18.07it/s] Batches: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████| 784/784 [00:08<00:00, 96.41it/s] Querying FAISS index: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████| 7/7 [00:06<00:00, 1.06it/s] Metric Positive Negative Difference Count 100,231 487,865 Mean 0.6866 0.4194 0.2752 Median 0.7010 0.4102 0.2760 Std 0.1125 0.0719 0.1136 Min 0.0303 0.1702 0.0209 25% 0.6221 0.3672 0.1899 50% 0.7010 0.4102 0.2760 75% 0.7667 0.4647 0.3590 Max 0.9584 0.7621 0.7073 Skipped 427,503 potential negatives (8.36%) due to the relative_margin of 0.05. Skipped 978 potential negatives (0.02%) due to the max_score of 0.8. Could not find enough negatives for 13290 samples (2.65%). Consider adjusting the range_max, range_min, relative_margin and max_score parameters if you'd like to find more valid negatives. >>> dataset Dataset({ features: ['query', 'answer', 'negative'], num_rows: 487865 }) >>> dataset[0] { 'query': 'when did richmond last play in a preliminary final', 'answer': "Richmond Football Club Richmond began 2017 with 5 straight wins, a feat it had not achieved since 1995. A series of close losses hampered the Tigers throughout the middle of the season, including a 5-point loss to the Western Bulldogs, 2-point loss to Fremantle, and a 3-point loss to the Giants. Richmond ended the season strongly with convincing victories over Fremantle and St Kilda in the final two rounds, elevating the club to 3rd on the ladder. Richmond's first final of the season against the Cats at the MCG attracted a record qualifying final crowd of 95,028; the Tigers won by 51 points. Having advanced to the first preliminary finals for the first time since 2001, Richmond defeated Greater Western Sydney by 36 points in front of a crowd of 94,258 to progress to the Grand Final against Adelaide, their first Grand Final appearance since 1982. The attendance was 100,021, the largest crowd to a grand final since 1986. The Crows led at quarter time and led by as many as 13, but the Tigers took over the game as it progressed and scored seven straight goals at one point. They eventually would win by 48 points – 16.12 (108) to Adelaide's 8.12 (60) – to end their 37-year flag drought.[22] Dustin Martin also became the first player to win a Premiership medal, the Brownlow Medal and the Norm Smith Medal in the same season, while Damien Hardwick was named AFL Coaches Association Coach of the Year. Richmond's jump from 13th to premiers also marked the biggest jump from one AFL season to the next.", 'negative': "2018 NRL Grand Final The 2018 NRL Grand Final was the conclusive and premiership-deciding game of the 2018 National Rugby League season and was played on Sunday September 30 at Sydney's ANZ Stadium.[1] The match was contested between minor premiers the Sydney Roosters and defending premiers the Melbourne Storm. In front of a crowd of 82,688, Sydney won the match 21–6 to claim their 14th premiership title and their first since 2013. Roosters five-eighth Luke Keary was awarded the Clive Churchill Medal as the game's official man of the match." } >>> dataset.push_to_hub("natural-questions-hard-negatives", "triplet-all") Args: dataset (Dataset): A dataset containing (anchor, positive) pairs. model (SentenceTransformer): A SentenceTransformer model to use for embedding the sentences. anchor_column_name (str, optional): The column name in `dataset` that contains the anchor/query. Defaults to None, in which case the first column in `dataset` will be used. positive_column_name (str, optional): The column name in `dataset` that contains the positive candidates. Defaults to None, in which case the second column in `dataset` will be used. corpus (List[str], optional): A list containing documents as strings that will be used as candidate negatives in addition to the second column in `dataset`. Defaults to None, in which case the second column in `dataset` will exclusively be used as the negative candidate corpus. cross_encoder (CrossEncoder, optional): A CrossEncoder model to use for rescoring the candidates. Defaults to None. range_min (int): Minimum rank of the closest matches to consider as negatives. Defaults to 0. range_max (int, optional): Maximum rank of the closest matches to consider as negatives. Defaults to None. max_score (float, optional): Maximum score to consider as a negative. Defaults to None. min_score (float, optional): Minimum score to consider as a negative. Defaults to None. absolute_margin (float, optional): Absolute margin for hard negative mining, i.e. the minimum distance between the positive similarity and the negative similarity. Defaults to None. relative_margin (float, optional): Relative margin for hard negative mining, i.e. the maximum ratio between the positive similarity and the negative similarity. A value of 0.05 means that the negative is at most 95% as similar to the anchor as the positive. Defaults to None. num_negatives (int): Number of negatives to sample. Defaults to 3. sampling_strategy (Literal["random", "top"]): Sampling strategy for negatives: "top" or "random". Defaults to "top". query_prompt_name (Optional[str], optional): The name of a predefined prompt to use when encoding the first/anchor dataset column. It must match a key in the ``model.prompts`` dictionary, which can be set during model initialization or loaded from the model configuration. For example, if ``query_prompt_name="query"`` and the model prompts dictionary includes {"query": "query: "}, then the sentence "What is the capital of France?" is transformed into: "query: What is the capital of France?" before encoding. This is useful for models that were trained or fine-tuned with specific prompt formats. Ignored if ``query_prompt`` is provided. Defaults to None. query_prompt (Optional[str], optional): A raw prompt string to prepend directly to the first/anchor dataset column during encoding. For instance, `query_prompt="query: "` transforms the sentence "What is the capital of France?" into: "query: What is the capital of France?". Use this to override the prompt logic entirely and supply your own prefix. This takes precedence over ``query_prompt_name``. Defaults to None. corpus_prompt_name (Optional[str], optional): The name of a predefined prompt to use when encoding the corpus. See ``query_prompt_name`` for more information. Defaults to None. corpus_prompt (Optional[str], optional): A raw prompt string to prepend directly to the corpus during encoding. See ``query_prompt`` for more information. Defaults to None. include_positives (bool): Whether to include the positives in the negative candidates. Setting this to True is primarily useful for creating Reranking evaluation datasets for CrossEncoder models, where it can be useful to get a full ranking (including the positives) from a first-stage retrieval model. Defaults to False. output_format (Literal["triplet", "n-tuple", "n-tuple-scores", "labeled-pair", "labeled-list"]): Output format for the `datasets.Dataset`. Options are: - "triplet": (anchor, positive, negative) triplets, i.e. 3 columns. Useful for e.g. :class:`~sentence_transformers.cross_encoder.losses.CachedMultipleNegativesRankingLoss`. - "n-tuple": (anchor, positive, negative_1, ..., negative_n) tuples, i.e. 2 + num_negatives columns. Useful for e.g. :class:`~sentence_transformers.cross_encoder.losses.CachedMultipleNegativesRankingLoss`. - "n-tuple-scores": (anchor, positive, negative_1, ..., negative_n, score) tuples, i.e. 2 + num_negatives columns, but with one score value that's a list of similarities for the query-positive and each of the query-negative pairs. Useful for e.g. :class:`~sentence_transformers.sparse_encoder.losses.SparseMarginMSELoss`. - "labeled-pair": (anchor, passage, label) text tuples with a label of 0 for negative and 1 for positive, i.e. 3 columns. Useful for e.g. :class:`~sentence_transformers.cross_encoder.losses.BinaryCrossEntropyLoss`. - "labeled-list": (anchor, [doc1, doc2, ..., docN], [label1, label2, ..., labelN]) triplets with labels of 0 for negative and 1 for positive, i.e. 3 columns. Useful for e.g. :class:`~sentence_transformers.cross_encoder.losses.LambdaLoss`. Defaults to "triplet". batch_size (int): Batch size for encoding the dataset. Defaults to 32. faiss_batch_size (int): Batch size for FAISS top-k search. Defaults to 16384. use_faiss (bool): Whether to use FAISS for similarity search. May be recommended for large datasets. Defaults to False. use_multi_process (bool | List[str], optional): Whether to use multi-GPU/CPU processing. If True, uses all GPUs if CUDA is available, and 4 CPU processes if it's not available. You can also pass a list of PyTorch devices like ["cuda:0", "cuda:1", ...] or ["cpu", "cpu", "cpu", "cpu"]. verbose (bool): Whether to print statistics and logging. Defaults to True. cache_folder (str, optional): Directory path for caching embeddings. If provided, the function will save ``query_embeddings_{hash}.npy`` and ``corpus_embeddings_{hash}.npy`` under this folder after the first run, and on subsequent calls will load from these files if they exist to avoid recomputation. The hashes are computed based on the model name and the queries/corpus. Defaults to None. as_triplets (bool, optional): Deprecated. Use `output_format` instead. Defaults to None. margin (float, optional): Deprecated. Use `absolute_margin` or `relative_margin` instead. Defaults to None. Returns: Dataset: A dataset containing (anchor, positive, negative) triplets, (anchor, passage, label) text tuples with a label, or (anchor, positive, negative_1, ..., negative_n) tuples. zGPlease install `datasets` to use this function: `pip install datasets`.rz9The dataset is empty. Please provide a non-empty dataset.r rz)Dataset must contain exactly two columns.Ntripletn-tuplezrThe `as_triplets` parameter is deprecated. Use the `output_format` parameter instead. Setting `output_format` to `z`.topzWhen using `include_positives=True`, updating `range_min`, `range_max`, `max_score`, `margin`, or `sampling_strategy` from the default values may still discard the positive values.z~When using `include_positives=True`, `output_format` will be set to `"n-tuple"` to ensure that the ranking order is preserved.)rrn-tuple-scores labeled-pair labeled-listzInvalid output_format: z[. Must be one of 'triplet', 'n-tuple', 'n-tuple-scores', 'labeled-pair', or 'labeled-list'.zThe `margin` parameter is deprecated. Use the `absolute_margin` and/or `relative_margin` parameter instead. Setting `absolute_margin` to ` iz\Using FAISS, we can only retrieve up to 2048 documents per query. Setting range_max to 2048.zSetting range_max to z" based on the provided parameters.zFound z unique queries out of z total queries.zFound an average of z.3fz positives per query.T)exist_okF)usedforsecurityquery_embeddings_z.npycorpus_embeddings_z%[Cache] Loaded query embeddings from z (shape=)z&[Cache] Loaded corpus embeddings from )target_devices)pool batch_sizenormalize_embeddingsconvert_to_numpyshow_progress_bar prompt_nameprompt)r r!r"r#r$r%z"[Cache] Saved query embeddings to z#[Cache] Saved corpus embeddings to )cozQuerying FAISS index)desc)k)axis)r(dim)devicezRescoring with CrossEncoder)r'total)r convert_to_tensorinf)r+dtype)skippedratioabsolute_marginrelative_margin max_score min_scorerandom)min)r* descendingz/Negative candidates mined, preparing dataset...negativerlabel)rr)r*)start negative_rscorerlabelszHNo triplets could be generated. Please check the parameters and dataset.z{:<6} {:>14} {:>14} {:>14}cbt|tjr|jdS|dS)Nz.4f,) isinstancetorchTensoritem)values |/var/www/html/ai-insurance-compliance-backend/venv/lib/python3.12/site-packages/sentence_transformers/util/hard_negatives.pyz%mine_hard_negatives..s)Juell<[ejjl3%7dijkclMetricPositiveNegative DifferenceCountmeanmedianstdr7ch|jdkDrtj|StdS)Nrr.)numelrBr7floatscoress rFrGz%mine_hard_negatives..s' 8J599V#4PUV[P\rHz25%c|jdkDr%tj|jdStdS)Nrg?qr.rRrBquantilerSrTs rFrGz%mine_hard_negatives..0V\\^^_M_5>>&,,.D#IejkpeqrHz50%c|jdkDr%tj|jdStdS)Nrg?rWr.rYrTs rFrGz%mine_hard_negatives..s0FLLN]^L^5>>&,,.C#HdijodprHz75%c|jdkDr%tj|jdStdS)Nrg?rWr.rYrTs rFrGz%mine_hard_negatives..r[rHmaxch|jdkDrtj|StdS)Nrz-inf)rRrBr^rSrTs rFrGz%mine_hard_negatives..s' 8J599V#4PUV\P]rHr0r1zSkipped r@z potential negatives (z.2%z ) due to the z of . range_max range_minz, z and z$Could not find enough negatives for z samples (z). Consider adjusting the z parametersz, if you'd like to find more valid negatives.)ar ImportErrorlen ValueErrordatasetsr column_namesloggerwarninglist to_pandasgroupbycountto_dictvaluesr^printdictfromkeys enumeratecopyrBarange unsqueezer+nprNosmakedirsmodel_card_data base_modelhashlibsha256joinencode hexdigestpathexistsloadshapestart_multi_process_poolrAboolstop_multi_process_poolsavefaiss IndexFlatIP get_sentence_embedding_dimensionGpuMultipleClonerOptionsshard useFloat16index_cpu_to_all_gpus Exceptionaddrsearchappend from_numpy concatenateto similaritytopkrangezipextendtensorcattolistsimilarity_pairwiserpredictstackisinrSrRemptyr/r7repeatsizeTsumrDisinfclampr6samplesort empty_likeallflattenany from_dictformatrOrP capitalize)sdatasetmodelanchor_column_namepositive_column_namecorpus cross_encoderrbrar4r5r2r3 num_negativessampling_strategyquery_prompt_name query_promptcorpus_prompt_name corpus_promptinclude_positives output_formatr faiss_batch_size use_faissuse_multi_processverbose cache_folder as_tripletsmarginr columnspositives_per_query max_positives log_countersqueries positivesseparate_corpusidxtext corpus_idx all_queriesquery queries_idx n_queries batch_idxr+avg_positives_per_querycorpus_embeddingsquery_embeddings model_name query_hash corpus_hashquery_cache_filecorpus_cache_filerrindexr& scores_list indices_listi query_chunkrUindices_positive_indicespositive query_idxp n_positivesdoc_idxpositive_embeddingspositive_scores candidate_idx_idxcandidate_passages pred_scoresq_idx positive_masknum_candidatesmax_positive_scores start_idxremoved_indices num_skippednegative_scores local_indices num_optionsoptions sampled_idxindices_to_keepanchor_indices pos_indices dataset_data anchor_idx positive_idx negative_idxdifference_scoresnegative_scorekeep neg_indicesneg_idxkeep_row indices_rowoutput_dataset row_format formattermetricfunction param_name param_valuer0r1missing_negatives solutionsconsiderationsmissing_negatives_ratioss rFmine_hard_negativesrs` ! "cdd 7|qTUU ""G !37!B$QZ #7w#F&qz &:s7|q?PDEE%0 i  ++8/ =  N$$! E) NNe  I % NNQ &Mdd%m_5P Q    --<,=R A ##$67==?GGIJ^_ffh+,M  &/*EI^!]R%78=HI"M1MAI t  Itu  ))4VW XL7-./GW123ID(O $-- 23 4F.7v->? T$)?J?,,.K4==)*G09'0BC*#u5#:CKCG I Y'11"5I \\FC $$ yk!8[9I8J/Z[q"$''*="> $%>* +!ww'78 =>N=OxXhXnXnWoopqr 77>>+ , "(9 : >?P>QQYZkZqZqYrrstu $4$< 11'12CT'JtPa2D!($)LL))-%)&* 2(%1 %! '#(<<))-%)&* 1'$0 $   ) )$ / ($)LL))-%)&* 2(%1%! '#(<<))-%)&* 1'$0$ ww~~./ GG$&6 7:;K:LMNww~~/0 GG%'8 9;>,Q#GHKKFS!!"24EFII&Q **Vy=/HaP %*)$45q55{I6Ax& #**:h+?@A$44a3q64K4IKY>9I#9NOg&/OPGCL>K,<<=>AQQ1 Qv6QQ(i8H(jRWXcdgXhRi(jQ(j(jk+EII6F,G,N,N,PQ//0@BUVYYZ`aO #'BiF["&y'9@]ehipeq"r & CCLE;H!I4&,!I !I'//S%IM24FGH%"&0K &F3K &(// [), -!"0   NST]N^ _UUZZ(8(? @ _ "'u }\\^N"o&A$kk)OIT_bpTp2q ./+-  &$':'A'A&++a.RS'T'V'VZ[^mZm'nnO',U|mF? #)--/446K>IT_bpTp2q ./+- 9,#(<-%))+002 &$~5)L %  9,#(<-%))+002 &$~5)L % &+ZZ)%K"O]i./G!YZ-()!YZ-8E!!^m^+,)!^m^*;< h &ll1o(=(=(?(C(CA(FF !''M': U`a'v}}U7^}Ea a)[01))[*@A)8)=)=!PT)=)U&)]23iiUS\M]^c,,[-=qA^_Gii]bcl]m nVY!5!"f '0 &C"A{A3!Mg&/!MM  , ,$)II 1;;B?QWX%fh  !*113+22=!DFFW__adss . ()eEl]: yQ`Ga tmc8emeqeqesS!1 t 4=c/SZ>[4\##0C0(K<<>3 SS^E_#hkdEcgF5M#hh# cXT\T`T`TbsaS3x=00c */:+22=!DFFWZii <Acdd&W&&|4N1 n  j*j,OP   #o./#o./   UZZ u|| $ EII  \ ] q r p q q r ] ^ !   FH !!%%'h78h78h'89:   ( 0  0 ) $ ) $ (  #J \)&z29=$Z09wqk)?c{-XbWccghsgttuv +S\9S=QQ q $ I1}  -*  !23*  !23$  -!YYy"~6N9~!'IbM"99&7=3w<;W&X # 67H6ITkloSpq++9*:*CPYN]^L^SdfDghTU  { @ DX   46 5 PR)k"J ``b_ nN!g"f!M !u#h# ds AO3 AO931AO? APAP AP AP2AP# AP)#AP. 'AP3AP8AP= AQ(AQ AQ AQ 2AQ @ AQ@AQ CAQCAQC?/AQ&D. AQ D9AQ EAQ&EAQ/E)AQ/O? AP P AP QAQ Q AQ&)NNNNrNNNNNrNNNNFr i@FFTNNN):rr rr r str | Nonerrrzlist[str] | NonerzCrossEncoder | Nonerbintraz int | Noner4 float | Noner5rr2rr3rrrrzLiteral['random', 'top']rrrrrrrrrrrzOLiteral['triplet', 'n-tuple', 'n-tuple-scores', 'labeled-pair', 'labeled-list']r rrrrrrzlist[str] | boolrrrrrz bool | Nonerrreturnr ) __future__rr}loggingryr6typingrrnumpyrxrBrrtqdm.autonotebook environmentr getLogger__name__rirgr 0sentence_transformers.cross_encoder.CrossEncoderr )sentence_transformers.SentenceTransformerr rrHrFr*s" ) ".   8 $ MM &*'+#)- ""$($(27$(#%) $#en!*/##9h  h  h #h % h   h  ' h h h h h "h "h h 0h "h  !h "##h $%h &'h (c)h *+h ,-h ./h 0(1h 23h 45h 67h 8 9h : ;h rH