--- tags: - mteb model-index: - name: Dmeta-embedding-zh-small results: - task: type: STS dataset: type: C-MTEB/AFQMC name: MTEB AFQMC config: default split: validation revision: None metrics: - type: cos_sim_pearson value: 55.38441014851534 - type: cos_sim_spearman value: 59.54284362578262 - type: euclidean_pearson value: 58.18592108890414 - type: euclidean_spearman value: 59.54284362133902 - type: manhattan_pearson value: 58.142197046175916 - type: manhattan_spearman value: 59.47943468645265 - task: type: STS dataset: type: C-MTEB/ATEC name: MTEB ATEC config: default split: test revision: None metrics: - type: cos_sim_pearson value: 55.96911621560259 - type: cos_sim_spearman value: 58.6334496101353 - type: euclidean_pearson value: 62.78426382809823 - type: euclidean_spearman value: 58.63344961011331 - type: manhattan_pearson value: 62.80625401678188 - type: manhattan_spearman value: 58.618722128260394 - task: type: Classification dataset: type: mteb/amazon_reviews_multi name: MTEB AmazonReviewsClassification (zh) config: zh split: test revision: 1399c76144fd37290681b995c656ef9b2e06e26d metrics: - type: accuracy value: 44.88 - type: f1 value: 42.739249460584375 - task: type: STS dataset: type: C-MTEB/BQ name: MTEB BQ config: default split: test revision: None metrics: - type: cos_sim_pearson value: 68.56815521242152 - type: cos_sim_spearman value: 70.30776353631751 - type: euclidean_pearson value: 69.10087719019191 - type: euclidean_spearman value: 70.30775660748148 - type: manhattan_pearson value: 69.0672710967445 - type: manhattan_spearman value: 70.31940638148254 - task: type: Clustering dataset: type: C-MTEB/CLSClusteringP2P name: MTEB CLSClusteringP2P config: default split: test revision: None metrics: - type: v_measure value: 40.7861976704356 - task: type: Clustering dataset: type: C-MTEB/CLSClusteringS2S name: MTEB CLSClusteringS2S config: default split: test revision: None metrics: - type: v_measure value: 38.43028280281822 - task: type: Reranking dataset: type: C-MTEB/CMedQAv1-reranking name: MTEB CMedQAv1 config: default split: test revision: None metrics: - type: map value: 86.78386695617407 - type: mrr value: 88.79857142857142 - task: type: Reranking dataset: type: C-MTEB/CMedQAv2-reranking name: MTEB CMedQAv2 config: default split: test revision: None metrics: - type: map value: 87.38582377194436 - type: mrr value: 89.17158730158731 - task: type: Retrieval dataset: type: C-MTEB/CmedqaRetrieval name: MTEB CmedqaRetrieval config: default split: dev revision: None metrics: - type: map_at_1 value: 23.746000000000002 - type: map_at_10 value: 35.952 - type: map_at_100 value: 37.946999999999996 - type: map_at_1000 value: 38.059 - type: map_at_3 value: 31.680999999999997 - type: map_at_5 value: 34.046 - type: mrr_at_1 value: 36.409000000000006 - type: mrr_at_10 value: 44.801 - type: mrr_at_100 value: 45.842 - type: mrr_at_1000 value: 45.885999999999996 - type: mrr_at_3 value: 42.081 - type: mrr_at_5 value: 43.613 - type: ndcg_at_1 value: 36.409000000000006 - type: ndcg_at_10 value: 42.687000000000005 - type: ndcg_at_100 value: 50.352 - type: ndcg_at_1000 value: 52.275000000000006 - type: ndcg_at_3 value: 37.113 - type: ndcg_at_5 value: 39.434000000000005 - type: precision_at_1 value: 36.409000000000006 - type: precision_at_10 value: 9.712 - type: precision_at_100 value: 1.584 - type: precision_at_1000 value: 0.182 - type: precision_at_3 value: 21.096999999999998 - type: precision_at_5 value: 15.498999999999999 - type: recall_at_1 value: 23.746000000000002 - type: recall_at_10 value: 53.596 - type: recall_at_100 value: 85.232 - type: recall_at_1000 value: 98.092 - type: recall_at_3 value: 37.226 - type: recall_at_5 value: 44.187 - task: type: PairClassification dataset: type: C-MTEB/CMNLI name: MTEB Cmnli config: default split: validation revision: None metrics: - type: cos_sim_accuracy value: 82.66987372218881 - type: cos_sim_ap value: 90.28715189799232 - type: cos_sim_f1 value: 84.108318049412 - type: cos_sim_precision value: 78.0849358974359 - type: cos_sim_recall value: 91.13864858545709 - type: dot_accuracy value: 82.66987372218881 - type: dot_ap value: 90.29346021403634 - type: dot_f1 value: 84.108318049412 - type: dot_precision value: 78.0849358974359 - type: dot_recall value: 91.13864858545709 - type: euclidean_accuracy value: 82.66987372218881 - type: euclidean_ap value: 90.28656734732074 - type: euclidean_f1 value: 84.108318049412 - type: euclidean_precision value: 78.0849358974359 - type: euclidean_recall value: 91.13864858545709 - type: manhattan_accuracy value: 82.70595309681299 - type: manhattan_ap value: 90.25413574022456 - type: manhattan_f1 value: 83.9924670433145 - type: manhattan_precision value: 79.81052631578947 - type: manhattan_recall value: 88.63689501987373 - type: max_accuracy value: 82.70595309681299 - type: max_ap value: 90.29346021403634 - type: max_f1 value: 84.108318049412 - task: type: Retrieval dataset: type: C-MTEB/CovidRetrieval name: MTEB CovidRetrieval config: default split: dev revision: None metrics: - type: map_at_1 value: 61.38 - type: map_at_10 value: 71.23 - type: map_at_100 value: 71.61800000000001 - type: map_at_1000 value: 71.63000000000001 - type: map_at_3 value: 69.31 - type: map_at_5 value: 70.403 - type: mrr_at_1 value: 61.538000000000004 - type: mrr_at_10 value: 71.28999999999999 - type: mrr_at_100 value: 71.666 - type: mrr_at_1000 value: 71.678 - type: mrr_at_3 value: 69.44200000000001 - type: mrr_at_5 value: 70.506 - type: ndcg_at_1 value: 61.538000000000004 - type: ndcg_at_10 value: 75.626 - type: ndcg_at_100 value: 77.449 - type: ndcg_at_1000 value: 77.73400000000001 - type: ndcg_at_3 value: 71.75200000000001 - type: ndcg_at_5 value: 73.695 - type: precision_at_1 value: 61.538000000000004 - type: precision_at_10 value: 9.009 - type: precision_at_100 value: 0.9860000000000001 - type: precision_at_1000 value: 0.101 - type: precision_at_3 value: 26.379 - type: precision_at_5 value: 16.797 - type: recall_at_1 value: 61.38 - type: recall_at_10 value: 89.199 - type: recall_at_100 value: 97.576 - type: recall_at_1000 value: 99.789 - type: recall_at_3 value: 78.635 - type: recall_at_5 value: 83.325 - task: type: Retrieval dataset: type: C-MTEB/DuRetrieval name: MTEB DuRetrieval config: default split: dev revision: None metrics: - type: map_at_1 value: 23.067 - type: map_at_10 value: 70.658 - type: map_at_100 value: 73.85300000000001 - type: map_at_1000 value: 73.925 - type: map_at_3 value: 48.391 - type: map_at_5 value: 61.172000000000004 - type: mrr_at_1 value: 83.1 - type: mrr_at_10 value: 88.214 - type: mrr_at_100 value: 88.298 - type: mrr_at_1000 value: 88.304 - type: mrr_at_3 value: 87.717 - type: mrr_at_5 value: 88.03699999999999 - type: ndcg_at_1 value: 83.1 - type: ndcg_at_10 value: 79.89 - type: ndcg_at_100 value: 83.829 - type: ndcg_at_1000 value: 84.577 - type: ndcg_at_3 value: 78.337 - type: ndcg_at_5 value: 77.224 - type: precision_at_1 value: 83.1 - type: precision_at_10 value: 38.934999999999995 - type: precision_at_100 value: 4.6690000000000005 - type: precision_at_1000 value: 0.484 - type: precision_at_3 value: 70.48299999999999 - type: precision_at_5 value: 59.68 - type: recall_at_1 value: 23.067 - type: recall_at_10 value: 81.702 - type: recall_at_100 value: 94.214 - type: recall_at_1000 value: 98.241 - type: recall_at_3 value: 51.538 - type: recall_at_5 value: 67.39 - task: type: Retrieval dataset: type: C-MTEB/EcomRetrieval name: MTEB EcomRetrieval config: default split: dev revision: None metrics: - type: map_at_1 value: 49.8 - type: map_at_10 value: 59.46399999999999 - type: map_at_100 value: 60.063 - type: map_at_1000 value: 60.08 - type: map_at_3 value: 56.833 - type: map_at_5 value: 58.438 - type: mrr_at_1 value: 49.8 - type: mrr_at_10 value: 59.46399999999999 - type: mrr_at_100 value: 60.063 - type: mrr_at_1000 value: 60.08 - type: mrr_at_3 value: 56.833 - type: mrr_at_5 value: 58.438 - type: ndcg_at_1 value: 49.8 - type: ndcg_at_10 value: 64.48 - type: ndcg_at_100 value: 67.314 - type: ndcg_at_1000 value: 67.745 - type: ndcg_at_3 value: 59.06400000000001 - type: ndcg_at_5 value: 61.973 - type: precision_at_1 value: 49.8 - type: precision_at_10 value: 8.04 - type: precision_at_100 value: 0.935 - type: precision_at_1000 value: 0.097 - type: precision_at_3 value: 21.833 - type: precision_at_5 value: 14.52 - type: recall_at_1 value: 49.8 - type: recall_at_10 value: 80.4 - type: recall_at_100 value: 93.5 - type: recall_at_1000 value: 96.8 - type: recall_at_3 value: 65.5 - type: recall_at_5 value: 72.6 - task: type: Classification dataset: type: C-MTEB/IFlyTek-classification name: MTEB IFlyTek config: default split: validation revision: None metrics: - type: accuracy value: 49.111196614082345 - type: f1 value: 37.07930546974089 - task: type: Classification dataset: type: C-MTEB/JDReview-classification name: MTEB JDReview config: default split: test revision: None metrics: - type: accuracy value: 85.57223264540339 - type: ap value: 53.30690968994808 - type: f1 value: 80.20587062271773 - task: type: STS dataset: type: C-MTEB/LCQMC name: MTEB LCQMC config: default split: test revision: None metrics: - type: cos_sim_pearson value: 73.03085269274996 - type: cos_sim_spearman value: 78.72837937949888 - type: euclidean_pearson value: 78.34911745798928 - type: euclidean_spearman value: 78.72838602779268 - type: manhattan_pearson value: 78.31833697617105 - type: manhattan_spearman value: 78.69603741566397 - task: type: Reranking dataset: type: C-MTEB/Mmarco-reranking name: MTEB MMarcoReranking config: default split: dev revision: None metrics: - type: map value: 27.391692468538416 - type: mrr value: 26.44682539682539 - task: type: Retrieval dataset: type: C-MTEB/MMarcoRetrieval name: MTEB MMarcoRetrieval config: default split: dev revision: None metrics: - type: map_at_1 value: 57.206999999999994 - type: map_at_10 value: 66.622 - type: map_at_100 value: 67.12700000000001 - type: map_at_1000 value: 67.145 - type: map_at_3 value: 64.587 - type: map_at_5 value: 65.827 - type: mrr_at_1 value: 59.312 - type: mrr_at_10 value: 67.387 - type: mrr_at_100 value: 67.836 - type: mrr_at_1000 value: 67.851 - type: mrr_at_3 value: 65.556 - type: mrr_at_5 value: 66.66 - type: ndcg_at_1 value: 59.312 - type: ndcg_at_10 value: 70.748 - type: ndcg_at_100 value: 73.076 - type: ndcg_at_1000 value: 73.559 - type: ndcg_at_3 value: 66.81200000000001 - type: ndcg_at_5 value: 68.92399999999999 - type: precision_at_1 value: 59.312 - type: precision_at_10 value: 8.798 - type: precision_at_100 value: 0.996 - type: precision_at_1000 value: 0.104 - type: precision_at_3 value: 25.487 - type: precision_at_5 value: 16.401 - type: recall_at_1 value: 57.206999999999994 - type: recall_at_10 value: 82.767 - type: recall_at_100 value: 93.449 - type: recall_at_1000 value: 97.262 - type: recall_at_3 value: 72.271 - type: recall_at_5 value: 77.291 - task: type: Classification dataset: type: mteb/amazon_massive_intent name: MTEB MassiveIntentClassification (zh-CN) config: zh-CN split: test revision: 31efe3c427b0bae9c22cbb560b8f15491cc6bed7 metrics: - type: accuracy value: 70.78345662407531 - type: f1 value: 68.35683436974351 - task: type: Classification dataset: type: mteb/amazon_massive_scenario name: MTEB MassiveScenarioClassification (zh-CN) config: zh-CN split: test revision: 7d571f92784cd94a019292a1f45445077d0ef634 metrics: - type: accuracy value: 73.16408876933423 - type: f1 value: 73.31484873459382 - task: type: Retrieval dataset: type: C-MTEB/MedicalRetrieval name: MTEB MedicalRetrieval config: default split: dev revision: None metrics: - type: map_at_1 value: 51.4 - type: map_at_10 value: 57.091 - type: map_at_100 value: 57.652 - type: map_at_1000 value: 57.703 - type: map_at_3 value: 55.733 - type: map_at_5 value: 56.363 - type: mrr_at_1 value: 51.7 - type: mrr_at_10 value: 57.243 - type: mrr_at_100 value: 57.80499999999999 - type: mrr_at_1000 value: 57.855999999999995 - type: mrr_at_3 value: 55.883 - type: mrr_at_5 value: 56.513000000000005 - type: ndcg_at_1 value: 51.4 - type: ndcg_at_10 value: 59.948 - type: ndcg_at_100 value: 63.064 - type: ndcg_at_1000 value: 64.523 - type: ndcg_at_3 value: 57.089999999999996 - type: ndcg_at_5 value: 58.214 - type: precision_at_1 value: 51.4 - type: precision_at_10 value: 6.9 - type: precision_at_100 value: 0.845 - type: precision_at_1000 value: 0.096 - type: precision_at_3 value: 20.333000000000002 - type: precision_at_5 value: 12.740000000000002 - type: recall_at_1 value: 51.4 - type: recall_at_10 value: 69.0 - type: recall_at_100 value: 84.5 - type: recall_at_1000 value: 96.2 - type: recall_at_3 value: 61.0 - type: recall_at_5 value: 63.7 - task: type: Classification dataset: type: C-MTEB/MultilingualSentiment-classification name: MTEB MultilingualSentiment config: default split: validation revision: None metrics: - type: accuracy value: 74.38999999999999 - type: f1 value: 74.07161306140839 - task: type: PairClassification dataset: type: C-MTEB/OCNLI name: MTEB Ocnli config: default split: validation revision: None metrics: - type: cos_sim_accuracy value: 81.15863562533838 - type: cos_sim_ap value: 84.84571607908443 - type: cos_sim_f1 value: 82.55872063968016 - type: cos_sim_precision value: 78.36812144212524 - type: cos_sim_recall value: 87.22280887011615 - type: dot_accuracy value: 81.15863562533838 - type: dot_ap value: 84.84571607908443 - type: dot_f1 value: 82.55872063968016 - type: dot_precision value: 78.36812144212524 - type: dot_recall value: 87.22280887011615 - type: euclidean_accuracy value: 81.15863562533838 - type: euclidean_ap value: 84.84571607908443 - type: euclidean_f1 value: 82.55872063968016 - type: euclidean_precision value: 78.36812144212524 - type: euclidean_recall value: 87.22280887011615 - type: manhattan_accuracy value: 80.7796426637791 - type: manhattan_ap value: 84.81524098914134 - type: manhattan_f1 value: 82.36462990561351 - type: manhattan_precision value: 77.76735459662288 - type: manhattan_recall value: 87.53959873284055 - type: max_accuracy value: 81.15863562533838 - type: max_ap value: 84.84571607908443 - type: max_f1 value: 82.55872063968016 - task: type: Classification dataset: type: C-MTEB/OnlineShopping-classification name: MTEB OnlineShopping config: default split: test revision: None metrics: - type: accuracy value: 93.12000000000002 - type: ap value: 91.0749103088623 - type: f1 value: 93.10837266607813 - task: type: STS dataset: type: C-MTEB/PAWSX name: MTEB PAWSX config: default split: test revision: None metrics: - type: cos_sim_pearson value: 38.5692290188029 - type: cos_sim_spearman value: 42.965264868554335 - type: euclidean_pearson value: 43.002526263615735 - type: euclidean_spearman value: 42.97561576045246 - type: manhattan_pearson value: 43.050089639788936 - type: manhattan_spearman value: 43.038497558804934 - task: type: STS dataset: type: C-MTEB/QBQTC name: MTEB QBQTC config: default split: test revision: None metrics: - type: cos_sim_pearson value: 38.99284895602663 - type: cos_sim_spearman value: 41.02655813481606 - type: euclidean_pearson value: 38.934953519378354 - type: euclidean_spearman value: 41.02680077136343 - type: manhattan_pearson value: 39.224809609807785 - type: manhattan_spearman value: 41.13950779185706 - task: type: STS dataset: type: mteb/sts22-crosslingual-sts name: MTEB STS22 (zh) config: zh split: test revision: 6d1ba47164174a496b7fa5d3569dae26a6813b80 metrics: - type: cos_sim_pearson value: 66.47464607633356 - type: cos_sim_spearman value: 66.76311382148693 - type: euclidean_pearson value: 67.25180409604143 - type: euclidean_spearman value: 66.76311382148693 - type: manhattan_pearson value: 67.6928257682864 - type: manhattan_spearman value: 67.08172581019826 - task: type: STS dataset: type: C-MTEB/STSB name: MTEB STSB config: default split: test revision: None metrics: - type: cos_sim_pearson value: 77.48943840585562 - type: cos_sim_spearman value: 79.0869194735025 - type: euclidean_pearson value: 79.48559575794792 - type: euclidean_spearman value: 79.08765044225807 - type: manhattan_pearson value: 79.36157224751007 - type: manhattan_spearman value: 78.94400905463999 - task: type: Reranking dataset: type: C-MTEB/T2Reranking name: MTEB T2Reranking config: default split: dev revision: None metrics: - type: map value: 66.1093201711458 - type: mrr value: 75.70959742506797 - task: type: Retrieval dataset: type: C-MTEB/T2Retrieval name: MTEB T2Retrieval config: default split: dev revision: None metrics: - type: map_at_1 value: 25.533 - type: map_at_10 value: 71.322 - type: map_at_100 value: 75.244 - type: map_at_1000 value: 75.333 - type: map_at_3 value: 50.15500000000001 - type: map_at_5 value: 61.514 - type: mrr_at_1 value: 86.126 - type: mrr_at_10 value: 89.462 - type: mrr_at_100 value: 89.58500000000001 - type: mrr_at_1000 value: 89.59 - type: mrr_at_3 value: 88.88000000000001 - type: mrr_at_5 value: 89.241 - type: ndcg_at_1 value: 86.126 - type: ndcg_at_10 value: 79.89500000000001 - type: ndcg_at_100 value: 84.405 - type: ndcg_at_1000 value: 85.286 - type: ndcg_at_3 value: 81.547 - type: ndcg_at_5 value: 79.834 - type: precision_at_1 value: 86.126 - type: precision_at_10 value: 39.972 - type: precision_at_100 value: 4.932 - type: precision_at_1000 value: 0.514 - type: precision_at_3 value: 71.49 - type: precision_at_5 value: 59.687 - type: recall_at_1 value: 25.533 - type: recall_at_10 value: 78.962 - type: recall_at_100 value: 93.413 - type: recall_at_1000 value: 97.89099999999999 - type: recall_at_3 value: 52.129000000000005 - type: recall_at_5 value: 65.444 - task: type: Classification dataset: type: C-MTEB/TNews-classification name: MTEB TNews config: default split: validation revision: None metrics: - type: accuracy value: 51.800000000000004 - type: f1 value: 50.07807183704828 - task: type: Clustering dataset: type: C-MTEB/ThuNewsClusteringP2P name: MTEB ThuNewsClusteringP2P config: default split: test revision: None metrics: - type: v_measure value: 65.15253218390774 - task: type: Clustering dataset: type: C-MTEB/ThuNewsClusteringS2S name: MTEB ThuNewsClusteringS2S config: default split: test revision: None metrics: - type: v_measure value: 58.81779372506517 - task: type: Retrieval dataset: type: C-MTEB/VideoRetrieval name: MTEB VideoRetrieval config: default split: dev revision: None metrics: - type: map_at_1 value: 53.0 - type: map_at_10 value: 63.422999999999995 - type: map_at_100 value: 63.995000000000005 - type: map_at_1000 value: 64.004 - type: map_at_3 value: 61.382999999999996 - type: map_at_5 value: 62.488 - type: mrr_at_1 value: 53.0 - type: mrr_at_10 value: 63.422999999999995 - type: mrr_at_100 value: 63.995000000000005 - type: mrr_at_1000 value: 64.004 - type: mrr_at_3 value: 61.382999999999996 - type: mrr_at_5 value: 62.488 - type: ndcg_at_1 value: 53.0 - type: ndcg_at_10 value: 68.301 - type: ndcg_at_100 value: 70.988 - type: ndcg_at_1000 value: 71.294 - type: ndcg_at_3 value: 64.11 - type: ndcg_at_5 value: 66.094 - type: precision_at_1 value: 53.0 - type: precision_at_10 value: 8.35 - type: precision_at_100 value: 0.958 - type: precision_at_1000 value: 0.098 - type: precision_at_3 value: 24.0 - type: precision_at_5 value: 15.36 - type: recall_at_1 value: 53.0 - type: recall_at_10 value: 83.5 - type: recall_at_100 value: 95.8 - type: recall_at_1000 value: 98.3 - type: recall_at_3 value: 72.0 - type: recall_at_5 value: 76.8 - task: type: Classification dataset: type: C-MTEB/waimai-classification name: MTEB Waimai config: default split: test revision: None metrics: - type: accuracy value: 86.18 - type: ap value: 69.04229346593745 - type: f1 value: 84.52986739717021 license: apache-2.0 ---
icon

Dmeta-embedding-small

- Dmeta-embedding系列模型是跨领域、跨任务、开箱即用的中文 Embedding 模型，适用于搜索、问答、智能客服、LLM+RAG 等各种业务场景，支持使用 Transformers/Sentence-Transformers/Langchain 等工具加载推理。 - **Dmeta-embedding-zh-small**是开源模型[Dmeta-embedding-zh](https://huggingface.co/DMetaSoul/Dmeta-embedding-zh)的蒸馏版本（8层BERT），模型大小不到300M。相较于原始版本，Dmeta-embedding-zh-small模型大小减小三分之一，推理速度提升约30%，总体精度下降约1.4%。 --- ## Evaluation 这里主要跟蒸馏前对应的 teacher 模型作了对比： *性能：*（基于1万条数据测试，GPU设备是V100） | | Teacher | Student | Gap | | ---------- | ------------------------- | ------------------------------ | ----- | | Model | Dmeta-Embedding-zh (411M) | Dmeta-Embedding-zh-small (297M)| 0.67x | | Cost | 127s | 89s | -30% | | Latency | 13ms | 9ms | -31% | | Throughput | 78 sentence/s | 111 sentence/s | 1.4x | *精度：*（参考自MTEB榜单） | | **Classification** | **Clustering** | **Pair Classification** | **Reranking** | **Retrieval** | **STS** | **Avg** | | -------------- | ----------------- | -------------- | ----------------------- | ------------- | ------------- | ------- | ------- | | **Teacher** | 70 | 50.96 | 88.92 | 67.17 | 70.41 | 64.89 | 67.51 | | **Student** | 69.89 | 50.8 | 87.57 | 66.92 | 67.7 | 62.13 | 66.1 | | **Gap** | -0.11 | -0.16 | -1.35 | -0.25 | -2.71 | -2.76 | -1.41 | ## Usage 目前模型支持通过 [Sentence-Transformers](#sentence-transformers), [Langchain](#langchain), [Huggingface Transformers](#huggingface-transformers) 等主流框架进行推理，具体用法参考各个框架的示例。 ### Sentence-Transformers Dmeta-embedding 模型支持通过 [sentence-transformers](https://www.SBERT.net) 来加载推理： ``` pip install -U sentence-transformers ``` ```python from sentence_transformers import SentenceTransformer texts1 = ["胡子长得太快怎么办？", "在香港哪里买手表好"] texts2 = ["胡子长得快怎么办？", "怎样使胡子不浓密！", "香港买手表哪里好", "在杭州手机到哪里买"] model = SentenceTransformer('DMetaSoul/Dmeta-embedding-zh-small') embs1 = model.encode(texts1, normalize_embeddings=True) embs2 = model.encode(texts2, normalize_embeddings=True) # 计算两两相似度 similarity = embs1 @ embs2.T print(similarity) # 获取 texts1[i] 对应的最相似 texts2[j] for i in range(len(texts1)): scores = [] for j in range(len(texts2)): scores.append([texts2[j], similarity[i][j]]) scores = sorted(scores, key=lambda x:x[1], reverse=True) print(f"查询文本：{texts1[i]}") for text2, score in scores: print(f"相似文本：{text2}，打分：{score}") print() ``` 示例输出如下： ``` 查询文本：胡子长得太快怎么办？ 相似文本：胡子长得快怎么办？，打分：0.965681254863739 相似文本：怎样使胡子不浓密！，打分：0.7353651523590088 相似文本：香港买手表哪里好，打分：0.24928246438503265 相似文本：在杭州手机到哪里买，打分：0.2038613110780716 查询文本：在香港哪里买手表好 相似文本：香港买手表哪里好，打分：0.9916468262672424 相似文本：在杭州手机到哪里买，打分：0.498248815536499 相似文本：胡子长得快怎么办？，打分：0.2424771636724472 相似文本：怎样使胡子不浓密！，打分：0.21715955436229706 ``` ### Langchain Dmeta-embedding 模型支持通过 LLM 工具框架 [langchain](https://www.langchain.com/) 来加载推理： ``` pip install -U langchain ``` ```python import torch import numpy as np from langchain.embeddings import HuggingFaceEmbeddings model_name = "DMetaSoul/Dmeta-embedding-zh-small" model_kwargs = {'device': 'cuda' if torch.cuda.is_available() else 'cpu'} encode_kwargs = {'normalize_embeddings': True} # set True to compute cosine similarity model = HuggingFaceEmbeddings( model_name=model_name, model_kwargs=model_kwargs, encode_kwargs=encode_kwargs, ) texts1 = ["胡子长得太快怎么办？", "在香港哪里买手表好"] texts2 = ["胡子长得快怎么办？", "怎样使胡子不浓密！", "香港买手表哪里好", "在杭州手机到哪里买"] embs1 = model.embed_documents(texts1) embs2 = model.embed_documents(texts2) embs1, embs2 = np.array(embs1), np.array(embs2) # 计算两两相似度 similarity = embs1 @ embs2.T print(similarity) # 获取 texts1[i] 对应的最相似 texts2[j] for i in range(len(texts1)): scores = [] for j in range(len(texts2)): scores.append([texts2[j], similarity[i][j]]) scores = sorted(scores, key=lambda x:x[1], reverse=True) print(f"查询文本：{texts1[i]}") for text2, score in scores: print(f"相似文本：{text2}，打分：{score}") print() ``` ### HuggingFace Transformers Dmeta-embedding 模型支持通过 [HuggingFace Transformers](https://huggingface.co/docs/transformers/index) 框架来加载推理： ``` pip install -U transformers ``` ```python import torch from transformers import AutoTokenizer, AutoModel def mean_pooling(model_output, attention_mask): token_embeddings = model_output[0] #First element of model_output contains all token embeddings input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float() return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9) def cls_pooling(model_output): return model_output[0][:, 0] texts1 = ["胡子长得太快怎么办？", "在香港哪里买手表好"] texts2 = ["胡子长得快怎么办？", "怎样使胡子不浓密！", "香港买手表哪里好", "在杭州手机到哪里买"] tokenizer = AutoTokenizer.from_pretrained('DMetaSoul/Dmeta-embedding-zh-small') model = AutoModel.from_pretrained('DMetaSoul/Dmeta-embedding-zh-small') model.eval() with torch.no_grad(): inputs1 = tokenizer(texts1, padding=True, truncation=True, return_tensors='pt') inputs2 = tokenizer(texts2, padding=True, truncation=True, return_tensors='pt') model_output1 = model(**inputs1) model_output2 = model(**inputs2) embs1, embs2 = cls_pooling(model_output1), cls_pooling(model_output2) embs1 = torch.nn.functional.normalize(embs1, p=2, dim=1).numpy() embs2 = torch.nn.functional.normalize(embs2, p=2, dim=1).numpy() # 计算两两相似度 similarity = embs1 @ embs2.T print(similarity) # 获取 texts1[i] 对应的最相似 texts2[j] for i in range(len(texts1)): scores = [] for j in range(len(texts2)): scores.append([texts2[j], similarity[i][j]]) scores = sorted(scores, key=lambda x:x[1], reverse=True) print(f"查询文本：{texts1[i]}") for text2, score in scores: print(f"相似文本：{text2}，打分：{score}") print() ``` ## Contact 您如果在使用过程中，遇到任何问题，欢迎前往[讨论区](https://huggingface.co/DMetaSoul/Dmeta-embedding-zh-small/discussions)建言献策。 您也可以联系我们：赵中昊 , 肖文斌 , 孙凯 同时我们也开通了微信群，可扫码加入我们（人数超200了，先加管理员再拉进群），一起共建 AIGC 技术生态！ ## License Dmeta-embedding 系列模型采用 Apache-2.0 License，开源模型可以进行免费商用私有部署。