モデル概要

FRIDA は、テキスト分類、感情分析、クラスタリング、リランキングなど、さまざまな自然言語処理タスクに適した多機能モデルです。

モデル特徴

マルチタスクサポート

分類、クラスタリング、リランキングなど、さまざまな自然言語処理タスクをサポートします。

高性能

特にタイトル分類や不適切コンテンツ分類タスクにおいて、複数のデータセットで優れたパフォーマンスを発揮します。

多言語サポート

主にロシア語をサポートしますが、多言語タスクでも一定の性能を発揮します。

モデル能力

テキスト分類

感情分析

クラスタリング

リランキング

マルチラベル分類

使用事例

コンテンツモデレーション

不適切コンテンツ検出

ヘイトスピーチや有害情報などの不適切コンテンツを検出・分類するために使用されます。

精度78.33%、F1スコア78.20%

感情分析

映画レビュー感情分類

映画レビューの感情傾向を分析するために使用されます。

精度70.47%、F1スコア65.84%

情報検索

MIRACLリランキング

情報検索におけるドキュメントリランキングタスクに使用されます。

NDCG@10 66.04%、MAP@10 60.21%

model-index:

name: FRIDA results:
- dataset: config: default name: MTEB CEDRClassification (default) revision: c0ba03d058e3e1b2f3fd20518875a4563dd12db4 split: test type: ai-forever/cedr-classification metrics:
  - type: accuracy value: 64.60148777895856
  - type: f1 value: 70.36630348039266
  - type: lrap value: 92.47290116896953
  - type: main_score value: 64.60148777895856 task: type: MultilabelClassification
- dataset: config: default name: MTEB GeoreviewClassification (default) revision: 3765c0d1de6b7d264bc459433c45e5a75513839c split: test type: ai-forever/georeview-classification metrics:
  - type: accuracy value: 57.70996093750001
  - type: f1 value: 53.18542982057098
  - type: f1_weighted value: 53.17663229582108
  - type: main_score value: 57.70996093750001 task: type: Classification
- dataset: config: default name: MTEB GeoreviewClusteringP2P (default) revision: 97a313c8fc85b47f13f33e7e9a95c1ad888c7fec split: test type: ai-forever/georeview-clustering-p2p metrics:
  - type: main_score value: 78.25468393043356
  - type: v_measure value: 78.25468393043356
  - type: v_measure_std value: 0.5094366871364238 task: type: Clustering
- dataset: config: default name: MTEB HeadlineClassification (default) revision: 2fe05ee6b5832cda29f2ef7aaad7b7fe6a3609eb split: test type: ai-forever/headline-classification metrics:
  - type: accuracy value: 89.0185546875
  - type: f1 value: 88.993933120612
  - type: f1_weighted value: 88.99276764225768
  - type: main_score value: 89.0185546875 task: type: Classification
- dataset: config: default name: MTEB InappropriatenessClassification (default) revision: 601651fdc45ef243751676e62dd7a19f491c0285 split: test type: ai-forever/inappropriateness-classification metrics:
  - type: accuracy value: 78.330078125
  - type: ap value: 73.17856750532495
  - type: ap_weighted value: 73.17856750532495
  - type: f1 value: 78.20169867599041
  - type: f1_weighted value: 78.20169867599041
  - type: main_score value: 78.330078125 task: type: Classification
- dataset: config: default name: MTEB KinopoiskClassification (default) revision: 5911f26666ac11af46cb9c6849d0dc80a378af24 split: test type: ai-forever/kinopoisk-sentiment-classification metrics:
  - type: accuracy value: 70.46666666666665
  - type: f1 value: 65.83951766538878
  - type: f1_weighted value: 65.83951766538878
  - type: main_score value: 70.46666666666665 task: type: Classification
- dataset: config: ru name: MTEB MIRACLReranking (ru) revision: 6d1962c527217f8927fca80f890f14f36b2802af split: dev type: miracl/mmteb-miracl-reranking metrics:
  - type: MAP@1(MIRACL) value: 39.023
  - type: MAP@10(MIRACL) value: 60.208
  - type: MAP@100(MIRACL) value: 61.672000000000004
  - type: MAP@1000(MIRACL) value: 61.672000000000004
  - type: MAP@20(MIRACL) value: 61.30799999999999
  - type: MAP@3(MIRACL) value: 53.33
  - type: MAP@5(MIRACL) value: 57.289
  - type: NDCG@1(MIRACL) value: 63.352
  - type: NDCG@10(MIRACL) value: 66.042
  - type: NDCG@100(MIRACL) value: 68.702
  - type: NDCG@1000(MIRACL) value: 68.702
  - type: NDCG@20(MIRACL) value: 67.768
  - type: NDCG@3(MIRACL) value: 61.925
  - type: NDCG@5(MIRACL) value: 63.327
  - type: P@1(MIRACL) value: 63.352
  - type: P@10(MIRACL) value: 16.512
  - type: P@100(MIRACL) value: 1.9529999999999998
  - type: P@1000(MIRACL) value: 0.19499999999999998
  - type: P@20(MIRACL) value: 9.13
  - type: P@3(MIRACL) value: 37.878
  - type: P@5(MIRACL) value: 27.586
  - type: Recall@1(MIRACL) value: 39.023
  - type: Recall@10(MIRACL) value: 72.35000000000001
  - type: Recall@100(MIRACL) value: 79.952
  - type: Recall@1000(MIRACL) value: 79.952
  - type: Recall@20(MIRACL) value: 76.828
  - type: Recall@3(MIRACL) value: 57.769999999999996
  - type: Recall@5(MIRACL) value: 64.91900000000001
  - type: main_score value: 66.042
  - type: nAUC_MAP@1000_diff1(MIRACL) value: 27.150388833033052
  - type: nAUC_MAP@1000_max(MIRACL) value: 55.15672274267081
  - type: nAUC_MAP@1000_std(MIRACL) value: 30.088939934575553
  - type: nAUC_MAP@100_diff1(MIRACL) value: 27.150388833033052
  - type: nAUC_MAP@100_max(MIRACL) value: 55.15672274267081
  - type: nAUC_MAP@100_std(MIRACL) value: 30.088939934575553
  - type: nAUC_MAP@10_diff1(MIRACL) value: 27.853691773641742
  - type: nAUC_MAP@10_max(MIRACL) value: 52.89390350055654
  - type: nAUC_MAP@10_std(MIRACL) value: 28.08732516551691
  - type: nAUC_MAP@1_diff1(MIRACL) value: 43.23179150244192
  - type: nAUC_MAP@1_max(MIRACL) value: 29.923943954188864
  - type: nAUC_MAP@1_std(MIRACL) value: 7.447084370195121
  - type: nAUC_MAP@20_diff1(MIRACL) value: 27.328384072311675
  - type: nAUC_MAP@20_max(MIRACL) value: 54.60286379835721
  - type: nAUC_MAP@20_std(MIRACL) value: 29.8084128980043
  - type: nAUC_MAP@3_diff1(MIRACL) value: 31.244971536944554
  - type: nAUC_MAP@3_max(MIRACL) value: 43.63984692803854
  - type: nAUC_MAP@3_std(MIRACL) value: 18.609234683765887
  - type: nAUC_MAP@5_diff1(MIRACL) value: 29.088760492638286
  - type: nAUC_MAP@5_max(MIRACL) value: 48.30474364461509
  - type: nAUC_MAP@5_std(MIRACL) value: 23.817514353844224
  - type: nAUC_NDCG@1000_diff1(MIRACL) value: 23.12754356408408
  - type: nAUC_NDCG@1000_max(MIRACL) value: 64.24894553363303
  - type: nAUC_NDCG@1000_std(MIRACL) value: 38.19318050598967
  - type: nAUC_NDCG@100_diff1(MIRACL) value: 23.12754356408408
  - type: nAUC_NDCG@100_max(MIRACL) value: 64.24894553363303
  - type: nAUC_NDCG@100_std(MIRACL) value: 38.19318050598967
  - type: nAUC_NDCG@10_diff1(MIRACL) value: 24.779856373697275
  - type: nAUC_NDCG@10_max(MIRACL) value: 60.4054459738118
  - type: nAUC_NDCG@10_std(MIRACL) value: 35.148950441182784
  - type: nAUC_NDCG@1_diff1(MIRACL) value: 35.605865569438556
  - type: nAUC_NDCG@1_max(MIRACL) value: 65.77787399715454
  - type: nAUC_NDCG@1_std(MIRACL) value: 34.34726892885082
  - type: nAUC_NDCG@20_diff1(MIRACL) value: 23.71231783125691
  - type: nAUC_NDCG@20_max(MIRACL) value: 62.89676599488004
  - type: nAUC_NDCG@20_std(MIRACL) value: 37.697052941884316
  - type: nAUC_NDCG@3_diff1(MIRACL) value: 26.109027741640865
  - type: nAUC_NDCG@3_max(MIRACL) value: 56.22356793638693
  - type: nAUC_NDCG@3_std(MIRACL) value: 29.9437568508688
  - type: nAUC_NDCG@5_diff1(MIRACL) value: 25.98644715327336
  - type: nAUC_NDCG@5_max(MIRACL) value: 56.25032008404774
  - type: nAUC_NDCG@5_std(MIRACL) value: 31.581899860862578
  - type: nAUC_P@1000_diff1(MIRACL) value: -18.29912787064644
  - type: nAUC_P@1000_max(MIRACL) value: 31.811344878776087
  - type: nAUC_P@1000_std(MIRACL) value: 30.163820183304914
  - type: nAUC_P@100_diff1(MIRACL) value: -18.299127870646405
  - type: nAUC_P@100_max(MIRACL) value: 31.811344878776133
  - type: nAUC_P@100_std(MIRACL) value: 30.163820183304956
  - type: nAUC_P@10_diff1(MIRACL) value: -15.96416268531149
  - type: nAUC_P@10_max(MIRACL) value: 36.989578896466526
  - type: nAUC_P@10_std(MIRACL) value: 34.54507111688143
  - type: nAUC_P@1_diff1(MIRACL) value: 35.605865569438556
  - type: nAUC_P@1_max(MIRACL) value: 65.77787399715454
  - type: nAUC_P@1_std(MIRACL) value: 34.34726892885082
  - type: nAUC_P@20_diff1(MIRACL) value: -17.443963421383287
  - type: nAUC_P@20_max(MIRACL) value: 34.309618168778385
  - type: nAUC_P@20_std(MIRACL) value: 33.38820956485373
  - type: nAUC_P@3_diff1(MIRACL) value: -8.533621861815652
  - type: nAUC_P@3_max(MIRACL) value: 45.90408386776497
  - type: nAUC_P@3_std(MIRACL) value: 34.50459351305535
  - type: nAUC_P@5_diff1(MIRACL) value: -13.207968899314865
  - type: nAUC_P@5_max(MIRACL) value: 40.37718282248973
  - type: nAUC_P@5_std(MIRACL) value: 35.601417332196206
  - type: nAUC_Recall@1000_diff1(MIRACL) value: 7.907304198177226
  - type: nAUC_Recall@1000_max(MIRACL) value: 77.82197832361145
  - type: nAUC_Recall@1000_std(MIRACL) value: 52.66957487246724
  - type: nAUC_Recall@100_diff1(MIRACL) value: 7.907304198177226
  - type: nAUC_Recall@100_max(MIRACL) value: 77.82197832361145
  - type: nAUC_Recall@100_std(MIRACL) value: 52.66957487246724
  - type: nAUC_Recall@10_diff1(MIRACL) value: 15.498121023488693
  - type: nAUC_Recall@10_max(MIRACL) value: 62.24320529338724
  - type: nAUC_Recall@10_std(MIRACL) value: 40.60221460946224
  - type: nAUC_Recall@1_diff1(MIRACL) value: 43.23179150244192
  - type: nAUC_Recall@1_max(MIRACL) value: 29.923943954188864
  - type: nAUC_Recall@1_std(MIRACL) value: 7.447084370195121
  - type: nAUC_Recall@20_diff1(MIRACL) value: 11.457044176116248
  - type: nAUC_Recall@20_max(MIRACL) value: 70.3493054342368
  - type: nAUC_Recall@20_std(MIRACL) value: 49.27124296325928
  - type: nAUC_Recall@3_diff1(MIRACL) value: 25.12077828977941
  - type: nAUC_Recall@3_max(MIRACL) value: 42.903379317937166
  - type: nAUC_Recall@3_std(MIRACL) value: 20.324501722161497
  - type: nAUC_Recall@5_diff1(MIRACL) value: 20.925701235197977
  - type: nAUC_Recall@5_max(MIRACL) value: 49.85323960390812
  - type: nAUC_Recall@5_std(MIRACL) value: 29.04484539530469 task: type: Reranking
- dataset: config: ru name: MTEB MIRACLRetrieval (ru) revision: main split: dev type: miracl/mmteb-miracl metrics:
  - type: main_score value: 71.882
  - type: map_at_1 value: 37.913000000000004
  - type: map_at_10 value: 62.604000000000006
  - type: map_at_100 value: 64.925
  - type: map_at_1000 value: 64.992
  - type: map_at_20 value: 64.081
  - type: map_at_3 value: 55.212
  - type: map_at_5 value: 59.445
  - type: mrr_at_1 value: 73.24281150159744
  - type: mrr_at_10 value: 81.65043866321825
  - type: mrr_at_100 value: 81.85391378818977
  - type: mrr_at_1000 value: 81.85753390802569
  - type: mrr_at_20 value: 81.81045606130179
  - type: mrr_at_3 value: 80.56443024494146
  - type: mrr_at_5 value: 81.30724174653893
  - type: nauc_map_at_1000_diff1 value: 26.962150235593356
  - type: nauc_map_at_1000_max value: 29.234958037854568
  - type: nauc_map_at_1000_std value: -2.4294465103633884
  - type: nauc_map_at_100_diff1 value: 26.92990252114163
  - type: nauc_map_at_100_max value: 29.206328533120118
  - type: nauc_map_at_100_std value: -2.437371090941197
  - type: nauc_map_at_10_diff1 value: 25.758265691179226
  - type: nauc_map_at_10_max value: 26.949978490795317
  - type: nauc_map_at_10_std value: -5.484961002106038
  - type: nauc_map_at_1_diff1 value: 34.70849461278043
  - type: nauc_map_at_1_max value: 12.778570893623042
  - type: nauc_map_at_1_std value: -13.018292652743938
  - type: nauc_map_at_20_diff1 value: 26.659923008218268
  - type: nauc_map_at_20_max value: 28.341440871568185
  - type: nauc_map_at_20_std value: -3.614549844913084
  - type: nauc_map_at_3_diff1 value: 27.197629021438203
  - type: nauc_map_at_3_max value: 20.701094874050856
  - type: nauc_map_at_3_std value: -12.062992301112041
  - type: nauc_map_at_5_diff1 value: 25.51793537203295
  - type: nauc_map_at_5_max value: 23.80396771243794
  - type: nauc_map_at_5_std value: -8.920465695323575
  - type: nauc_mrr_at_1000_diff1 value: 45.14819989592967
  - type: nauc_mrr_at_1000_max value: 53.29202156141053
  - type: nauc_mrr_at_1000_std value: 18.037336462510524
  - type: nauc_mrr_at_100_diff1 value: 45.15287600228451
  - type: nauc_mrr_at_100_max value: 53.29979751928615
  - type: nauc_mrr_at_100_std value: 18.04996604778386
  - type: nauc_mrr_at_10_diff1 value: 44.96865105944474
  - type: nauc_mrr_at_10_max value: 53.53323465323092
  - type: nauc_mrr_at_10_std value: 18.25001344917689
  - type: nauc_mrr_at_1_diff1 value: 46.16604946873163
  - type: nauc_mrr_at_1_max value: 48.573651103547874
  - type: nauc_mrr_at_1_std value: 13.764871626330915
  - type: nauc_mrr_at_20_diff1 value: 45.11925458479102
  - type: nauc_mrr_at_20_max value: 53.35685123898342
  - type: nauc_mrr_at_20_std value: 18.127344968819905
  - type: nauc_mrr_at_3_diff1 value: 45.377195452730234
  - type: nauc_mrr_at_3_max value: 53.35146309217089
  - type: nauc_mrr_at_3_std value: 17.47105877186237
  - type: nauc_mrr_at_5_diff1 value: 45.00525578771549
  - type: nauc_mrr_at_5_max value: 53.76227254707128
  - type: nauc_mrr_at_5_std value: 18.437290060746957
  - type: nauc_ndcg_at_1000_diff1 value: 31.19215594457491
  - type: nauc_ndcg_at_1000_max value: 38.09555406458668
  - type: nauc_ndcg_at_1000_std value: 7.225628621238009
  - type: nauc_ndcg_at_100_diff1 value: 30.726331247999934
  - type: nauc_ndcg_at_100_max value: 37.81369589418277
  - type: nauc_ndcg_at_100_std value: 7.242855238555071
  - type: nauc_ndcg_at_10_diff1 value: 27.514048333744835
  - type: nauc_ndcg_at_10_max value: 33.10990399385253
  - type: nauc_ndcg_at_10_std value: 0.3051899572112002
  - type: nauc_ndcg_at_1_diff1 value: 47.06089085235751
  - type: nauc_ndcg_at_1_max value: 47.7300872370495
  - type: nauc_ndcg_at_1_std value: 12.468605493613916
  - type: nauc_ndcg_at_20_diff1 value: 29.404215438764496
  - type: nauc_ndcg_at_20_max value: 35.26967886796471
  - type: nauc_ndcg_at_20_std value: 3.7214697890813353
  - type: nauc_ndcg_at_3_diff1 value: 29.448848639643067
  - type: nauc_ndcg_at_3_max value: 33.85912412370657
  - type: nauc_ndcg_at_3_std value: 0.895453646819452
  - type: nauc_ndcg_at_5_diff1 value: 26.916649012613526
  - type: nauc_ndcg_at_5_max value: 30.899005979291644
  - type: nauc_ndcg_at_5_std value: -1.0001575639156615
  - type: nauc_precision_at_1000_diff1 value: -8.492004667432635
  - type: nauc_precision_at_1000_max value: 14.970190384017679
  - type: nauc_precision_at_1000_std value: 32.871386621137816
  - type: nauc_precision_at_100_diff1 value: -8.287314133999967
  - type: nauc_precision_at_100_max value: 17.794821961284736
  - type: nauc_precision_at_100_std value: 35.092483550562
  - type: nauc_precision_at_10_diff1 value: -7.594128993028063
  - type: nauc_precision_at_10_max value: 24.691446370325732
  - type: nauc_precision_at_10_std value: 30.126552282608493
  - type: nauc_precision_at_1_diff1 value: 47.06089085235751
  - type: nauc_precision_at_1_max value: 47.7300872370495
  - type: nauc_precision_at_1_std value: 12.468605493613916
  - type: nauc_precision_at_20_diff1 value: -6.503872195775146
  - type: nauc_precision_at_20_max value: 21.789730053141312
  - type: nauc_precision_at_20_std value: 32.61349377558794
  - type: nauc_precision_at_3_diff1 value: 0.67417079971061
  - type: nauc_precision_at_3_max value: 30.793871354370662
  - type: nauc_precision_at_3_std value: 18.35266479252011
  - type: nauc_precision_at_5_diff1 value: -7.088881730215777
  - type: nauc_precision_at_5_max value: 26.539771712769006
  - type: nauc_precision_at_5_std value: 24.116262291865834
  - type: nauc_recall_at_1000_diff1 value: 34.53263588412461
  - type: nauc_recall_at_1000_max value: 63.54157869100173
  - type: nauc_recall_at_1000_std value: 64.19854844792808
  - type: nauc_recall_at_100_diff1 value: 22.86564728642275
  - type: nauc_recall_at_100_max value: 40.350507162549825
  - type: nauc_recall_at_100_std value: 29.24492545863015
  - type: nauc_recall_at_10_diff1 value: 15.384818367225009
  - type: nauc_recall_at_10_max value: 24.41108571453699
  - type: nauc_recall_at_10_std value: -3.9216160585776323
  - type: nauc_recall_at_1_diff1 value: 34.70849461278043
  - type: nauc_recall_at_1_max value: 12.778570893623042
  - type: nauc_recall_at_1_std value: -13.018292652743938
  - type: nauc_recall_at_20_diff1 value: 18.122499000084208
  - type: nauc_recall_at_20_max value: 26.63104220179424
  - type: nauc_recall_at_20_std value: 3.969217732521512
  - type: nauc_recall_at_3_diff1 value: 21.413050725250116
  - type: nauc_recall_at_3_max value: 16.18894988386887
  - type: nauc_recall_at_3_std value: -15.24884339282375
  - type: nauc_recall_at_5_diff1 value: 16.35673072212927
  - type: nauc_recall_at_5_max value: 18.607003829267846
  - type: nauc_recall_at_5_std value: -10.463525876945454
  - type: ndcg_at_1 value: 72.923
  - type: ndcg_at_10 value: 71.882
  - type: ndcg_at_100 value: 77.09899999999999
  - type: ndcg_at_1000 value: 77.835
  - type: ndcg_at_20 value: 74.497
  - type: ndcg_at_3 value: 68.504
  - type: ndcg_at_5 value: 69.068
  - type: precision_at_1 value: 72.923
  - type: precision_at_10 value: 19.936
  - type: precision_at_100 value: 2.6310000000000002
  - type: precision_at_1000 value: 0.27799999999999997
  - type: precision_at_20 value: 11.33
  - type: precision_at_3 value: 45.927
  - type: precision_at_5 value: 33.131
  - type: recall_at_1 value: 37.913000000000004
  - type: recall_at_10 value: 78.365
  - type: recall_at_100 value: 94.348
  - type: recall_at_1000 value: 98.187
  - type: recall_at_20 value: 85.229
  - type: recall_at_3 value: 61.42999999999999
  - type: recall_at_5 value: 69.56700000000001 task: type: Retrieval
- dataset: config: ru name: MTEB MassiveIntentClassification (ru) revision: 4672e20407010da34463acc759c162ca9734bca6 split: test type: mteb/amazon_massive_intent metrics:
  - type: accuracy value: 79.11903160726294
  - type: f1 value: 76.22609082694545
  - type: f1_weighted value: 77.81461248063566
  - type: main_score value: 79.11903160726294 task: type: Classification
- dataset: config: ru name: MTEB MassiveScenarioClassification (ru) revision: fad2c6e8459f9e1c45d9315f4953d921437d70f8 split: test type: mteb/amazon_massive_scenario metrics:
  - type: accuracy value: 88.80632145258912
  - type: f1 value: 87.53157475314829
  - type: f1_weighted value: 88.22733432521495
  - type: main_score value: 88.80632145258912 task: type: Classification
- dataset: config: default name: MTEB RUParaPhraserSTS (default) revision: 43265056790b8f7c59e0139acb4be0a8dad2c8f4 split: test type: merionum/ru_paraphraser metrics:
  - type: cosine_pearson value: 72.70307124858925
  - type: cosine_spearman value: 78.09439086920204
  - type: euclidean_pearson value: 76.2033672014715
  - type: euclidean_spearman value: 78.09439086920204
  - type: main_score value: 78.09439086920204
  - type: manhattan_pearson value: 76.11750470223116
  - type: manhattan_spearman value: 78.01081063503413
  - type: pearson value: 72.70307124858925
  - type: spearman value: 78.09439086920204 task: type: STS
- dataset: config: default name: MTEB RiaNewsRetrieval (default) revision: 82374b0bbacda6114f39ff9c5b925fa1512ca5d7 split: test type: ai-forever/ria-news-retrieval metrics:
  - type: main_score value: 86.819
  - type: map_at_1 value: 78.79
  - type: map_at_10 value: 84.516
  - type: map_at_100 value: 84.68
  - type: map_at_1000 value: 84.685
  - type: map_at_20 value: 84.624
  - type: map_at_3 value: 83.722
  - type: map_at_5 value: 84.246
  - type: mrr_at_1 value: 78.78
  - type: mrr_at_10 value: 84.51815476190441
  - type: mrr_at_100 value: 84.68390840473289
  - type: mrr_at_1000 value: 84.68947095200002
  - type: mrr_at_20 value: 84.62958130822527
  - type: mrr_at_3 value: 83.74499999999964
  - type: mrr_at_5 value: 84.23849999999955
  - type: nauc_map_at_1000_diff1 value: 82.09914867708899
  - type: nauc_map_at_1000_max value: 43.02024854784386
  - type: nauc_map_at_1000_std value: -22.919695880762777
  - type: nauc_map_at_100_diff1 value: 82.09705922783733
  - type: nauc_map_at_100_max value: 43.02697379581718
  - type: nauc_map_at_100_std value: -22.90719212899522
  - type: nauc_map_at_10_diff1 value: 82.04404594672894
  - type: nauc_map_at_10_max value: 43.06752103182731
  - type: nauc_map_at_10_std value: -23.007870153273576
  - type: nauc_map_at_1_diff1 value: 83.89134152210333
  - type: nauc_map_at_1_max value: 38.083626428503415
  - type: nauc_map_at_1_std value: -25.817960401194252
  - type: nauc_map_at_20_diff1 value: 82.08534662247806
  - type: nauc_map_at_20_max value: 43.074305042312346
  - type: nauc_map_at_20_std value: -22.91785703613217
  - type: nauc_map_at_3_diff1 value: 81.7967508697558
  - type: nauc_map_at_3_max value: 42.90927479098251
  - type: nauc_map_at_3_std value: -24.01312203859392
  - type: nauc_map_at_5_diff1 value: 81.90704517505098
  - type: nauc_map_at_5_max value: 43.05204677044616
  - type: nauc_map_at_5_std value: -23.267331507554896
  - type: nauc_mrr_at_1000_diff1 value: 82.11902348082472
  - type: nauc_mrr_at_1000_max value: 43.04118936353063
  - type: nauc_mrr_at_1000_std value: -22.858804296830773
  - type: nauc_mrr_at_100_diff1 value: 82.11685562002263
  - type: nauc_mrr_at_100_max value: 43.0482537895494
  - type: nauc_mrr_at_100_std value: -22.84431127787993
  - type: nauc_mrr_at_10_diff1 value: 82.06909958688058
  - type: nauc_mrr_at_10_max value: 43.07921689466605
  - type: nauc_mrr_at_10_std value: -22.957623576663234
  - type: nauc_mrr_at_1_diff1 value: 83.91147637794326
  - type: nauc_mrr_at_1_max value: 37.91917159543152
  - type: nauc_mrr_at_1_std value: -26.141868289283266
  - type: nauc_mrr_at_20_diff1 value: 82.10314004731809
  - type: nauc_mrr_at_20_max value: 43.09295406509764
  - type: nauc_mrr_at_20_std value: -22.862091782178787
  - type: nauc_mrr_at_3_diff1 value: 81.82117067269036
  - type: nauc_mrr_at_3_max value: 42.94628953323521
  - type: nauc_mrr_at_3_std value: -23.852510312400714
  - type: nauc_mrr_at_5_diff1 value: 81.92857441701598
  - type: nauc_mrr_at_5_max value: 43.129719354492934
  - type: nauc_mrr_at_5_std value: -23.145342272624085
  - type: nauc_ndcg_at_1000_diff1 value: 81.75015729717991
  - type: nauc_ndcg_at_1000_max value: 44.7266586308995
  - type: nauc_ndcg_at_1000_std value: -20.60663899715267
  - type: nauc_ndcg_at_100_diff1 value: 81.6897808298767
  - type: nauc_ndcg_at_100_max value: 44.99492791287099
  - type: nauc_ndcg_at_100_std value: -20.09637266506936
  - type: nauc_ndcg_at_10_diff1 value: 81.46290312197337
  - type: nauc_ndcg_at_10_max value: 45.30218378452244
  - type: nauc_ndcg_at_10_std value: -20.70393523891777
  - type: nauc_ndcg_at_1_diff1 value: 83.89134152210333
  - type: nauc_ndcg_at_1_max value: 38.083626428503415
  - type: nauc_ndcg_at_1_std value: -25.817960401194252
  - type: nauc_ndcg_at_20_diff1 value: 81.61080772657213
  - type: nauc_ndcg_at_20_max value: 45.36571800492172
  - type: nauc_ndcg_at_20_std value: -20.278763852504042
  - type: nauc_ndcg_at_3_diff1 value: 80.95965359410461
  - type: nauc_ndcg_at_3_max value: 44.756971949205834
  - type: nauc_ndcg_at_3_std value: -23.07797617717319
  - type: nauc_ndcg_at_5_diff1 value: 81.12417712163976
  - type: nauc_ndcg_at_5_max value: 45.15727381406512
  - type: nauc_ndcg_at_5_std value: -21.52861766165519
  - type: nauc_precision_at_1000_diff1 value: 76.80566850396093
  - type: nauc_precision_at_1000_max value: 82.45685370922442
  - type: nauc_precision_at_1000_std value: 46.93570976777808
  - type: nauc_precision_at_100_diff1 value: 77.21645520953484
  - type: nauc_precision_at_100_max value: 73.43604108309935
  - type: nauc_precision_at_100_std value: 31.978176891671367
  - type: nauc_precision_at_10_diff1 value: 77.88251664302092
  - type: nauc_precision_at_10_max value: 60.58112638995018
  - type: nauc_precision_at_10_std value: -3.674424315180332
  - type: nauc_precision_at_1_diff1 value: 83.89134152210333
  - type: nauc_precision_at_1_max value: 38.083626428503415
  - type: nauc_precision_at_1_std value: -25.817960401194252
  - type: nauc_precision_at_20_diff1 value: 78.16426786697438
  - type: nauc_precision_at_20_max value: 66.0723612699222
  - type: nauc_precision_at_20_std value: 6.121527084555938
  - type: nauc_precision_at_3_diff1 value: 77.43122492166451
  - type: nauc_precision_at_3_max value: 52.50727288548085
  - type: nauc_precision_at_3_std value: -19.036076920799427
  - type: nauc_precision_at_5_diff1 value: 77.1127254320532
  - type: nauc_precision_at_5_max value: 56.100901899221135
  - type: nauc_precision_at_5_std value: -12.009191140844198
  - type: nauc_recall_at_1000_diff1 value: 76.80566850396035
  - type: nauc_recall_at_1000_max value: 82.45685370922577
  - type: nauc_recall_at_1000_std value: 46.93570976777776
  - type: nauc_recall_at_100_diff1 value: 77.21645520953459
  - type: nauc_recall_at_100_max value: 73.43604108310011
  - type: nauc_recall_at_100_std value: 31.978176891671993
  - type: nauc_recall_at_10_diff1 value: 77.88251664302089
  - type: nauc_recall_at_10_max value: 60.58112638994999
  - type: nauc_recall_at_10_std value: -3.6744243151805427
  - type: nauc_recall_at_1_diff1 value: 83.89134152210333
  - type: nauc_recall_at_1_max value: 38.083626428503415
  - type: nauc_recall_at_1_std value: -25.817960401194252
  - type: nauc_recall_at_20_diff1 value: 78.16426786697409
  - type: nauc_recall_at_20_max value: 66.07236126992217
  - type: nauc_recall_at_20_std value: 6.121527084555941
  - type: nauc_recall_at_3_diff1 value: 77.43122492166454
  - type: nauc_recall_at_3_max value: 52.507272885480816
  - type: nauc_recall_at_3_std value: -19.036076920799776
  - type: nauc_recall_at_5_diff1 value: 77.11272543205318
  - type: nauc_recall_at_5_max value: 56.10090189922128
  - type: nauc_recall_at_5_std value: -12.009191140843809
  - type: ndcg_at_1 value: 78.79
  - type: ndcg_at_10 value: 86.819
  - type: ndcg_at_100 value: 87.599
  - type: ndcg_at_1000 value: 87.761
  - type: ndcg_at_20 value: 87.208
  - type: ndcg_at_3 value: 85.222
  - type: ndcg_at_5 value: 86.164
  - type: precision_at_1 value: 78.79
  - type: precision_at_10 value: 9.384
  - type: precision_at_100 value: 0.975
  - type: precision_at_1000 value: 0.099
  - type: precision_at_20 value: 4.769
  - type: precision_at_3 value: 29.842999999999996
  - type: precision_at_5 value: 18.362000000000002
  - type: recall_at_1 value: 78.79
  - type: recall_at_10 value: 93.84
  - type: recall_at_100 value: 97.45
  - type: recall_at_1000 value: 98.76
  - type: recall_at_20 value: 95.37
  - type: recall_at_3 value: 89.53
  - type: recall_at_5 value: 91.81 task: type: Retrieval
- dataset: config: default name: MTEB RuBQReranking (default) revision: 2e96b8f098fa4b0950fc58eacadeb31c0d0c7fa2 split: test type: ai-forever/rubq-reranking metrics:
  - type: main_score value: 77.07394404835635
  - type: map value: 77.07394404835635
  - type: mrr value: 82.53144412718882
  - type: nAUC_map_diff1 value: 45.29805217456628
  - type: nAUC_map_max value: 34.39894042439188
  - type: nAUC_map_std value: 21.11309674418275
  - type: nAUC_mrr_diff1 value: 54.783994737367046
  - type: nAUC_mrr_max value: 45.68526733900048
  - type: nAUC_mrr_std value: 28.22466385500339 task: type: Reranking
- dataset: config: default name: MTEB RuBQRetrieval (default) revision: e19b6ffa60b3bc248e0b41f4cc37c26a55c2a67b split: test type: ai-forever/rubq-retrieval metrics:
  - type: main_score value: 72.392
  - type: map_at_1 value: 47.370000000000005
  - type: map_at_10 value: 65.503
  - type: map_at_100 value: 66.38
  - type: map_at_1000 value: 66.42099999999999
  - type: map_at_20 value: 66.071
  - type: map_at_3 value: 61.439
  - type: map_at_5 value: 63.922999999999995
  - type: mrr_at_1 value: 67.37588652482269
  - type: mrr_at_10 value: 76.0066747345116
  - type: mrr_at_100 value: 76.25754138969413
  - type: mrr_at_1000 value: 76.26968825657428
  - type: mrr_at_20 value: 76.17548265904622
  - type: mrr_at_3 value: 74.61583924349881
  - type: mrr_at_5 value: 75.46690307328608
  - type: nauc_map_at_1000_diff1 value: 42.52570720187294
  - type: nauc_map_at_1000_max value: 37.40318318724238
  - type: nauc_map_at_1000_std value: 0.6037788201535506
  - type: nauc_map_at_100_diff1 value: 42.493410029691226
  - type: nauc_map_at_100_max value: 37.39802489244377
  - type: nauc_map_at_100_std value: 0.6071359951887154
  - type: nauc_map_at_10_diff1 value: 42.09833519659916
  - type: nauc_map_at_10_max value: 37.1184138958874
  - type: nauc_map_at_10_std value: 0.4063543094010351
  - type: nauc_map_at_1_diff1 value: 49.56605205141156
  - type: nauc_map_at_1_max value: 26.251096698710384
  - type: nauc_map_at_1_std value: -4.580748485387834
  - type: nauc_map_at_20_diff1 value: 42.33372393482018
  - type: nauc_map_at_20_max value: 37.416955604649985
  - type: nauc_map_at_20_std value: 0.6050577802787294
  - type: nauc_map_at_3_diff1 value: 42.362234475441845
  - type: nauc_map_at_3_max value: 34.56001379838821
  - type: nauc_map_at_3_std value: -1.507636598929042
  - type: nauc_map_at_5_diff1 value: 42.0202264882535
  - type: nauc_map_at_5_max value: 36.64306050200848
  - type: nauc_map_at_5_std value: -0.09509025708798424
  - type: nauc_mrr_at_1000_diff1 value: 58.99601742026931
  - type: nauc_mrr_at_1000_max value: 49.61561872452777
  - type: nauc_mrr_at_1000_std value: 2.3956102974352356
  - type: nauc_mrr_at_100_diff1 value: 58.9865943101085
  - type: nauc_mrr_at_100_max value: 49.6248111507265
  - type: nauc_mrr_at_100_std value: 2.411155095066369
  - type: nauc_mrr_at_10_diff1 value: 58.81758131092919
  - type: nauc_mrr_at_10_max value: 49.780365572616695
  - type: nauc_mrr_at_10_std value: 2.7068696565195944
  - type: nauc_mrr_at_1_diff1 value: 61.67036882487055
  - type: nauc_mrr_at_1_max value: 45.455271042821714
  - type: nauc_mrr_at_1_std value: -0.9370526815458349
  - type: nauc_mrr_at_20_diff1 value: 58.93674818203478
  - type: nauc_mrr_at_20_max value: 49.703218108625215
  - type: nauc_mrr_at_20_std value: 2.4473106598190415
  - type: nauc_mrr_at_3_diff1 value: 59.046856598788445
  - type: nauc_mrr_at_3_max value: 49.37161726123392
  - type: nauc_mrr_at_3_std value: 1.5110936686701506
  - type: nauc_mrr_at_5_diff1 value: 58.92289378915668
  - type: nauc_mrr_at_5_max value: 49.847638994134144
  - type: nauc_mrr_at_5_std value: 2.420421880131702
  - type: nauc_ndcg_at_1000_diff1 value: 45.56062215161734
  - type: nauc_ndcg_at_1000_max value: 41.507152286702
  - type: nauc_ndcg_at_1000_std value: 2.79388283208751
  - type: nauc_ndcg_at_100_diff1 value: 44.84064192570408
  - type: nauc_ndcg_at_100_max value: 41.50353573562353
  - type: nauc_ndcg_at_100_std value: 3.1804999773629357
  - type: nauc_ndcg_at_10_diff1 value: 43.341482144213614
  - type: nauc_ndcg_at_10_max value: 41.159590898395074
  - type: nauc_ndcg_at_10_std value: 2.945242338240843
  - type: nauc_ndcg_at_1_diff1 value: 62.23623985611396
  - type: nauc_ndcg_at_1_max value: 45.04945770947091
  - type: nauc_ndcg_at_1_std value: -0.8804967656575725
  - type: nauc_ndcg_at_20_diff1 value: 43.905372612093664
  - type: nauc_ndcg_at_20_max value: 41.797709837872446
  - type: nauc_ndcg_at_20_std value: 3.1853356915569653
  - type: nauc_ndcg_at_3_diff1 value: 44.18163998834299
  - type: nauc_ndcg_at_3_max value: 38.352891017864636
  - type: nauc_ndcg_at_3_std value: -0.8235767021150929
  - type: nauc_ndcg_at_5_diff1 value: 43.41374688421302
  - type: nauc_ndcg_at_5_max value: 40.390365601593956
  - type: nauc_ndcg_at_5_std value: 1.6743650108127537
  - type: nauc_precision_at_1000_diff1 value: -9.711058370691381
  - type: nauc_precision_at_1000_max value: 6.97321343449286
  - type: nauc_precision_at_1000_std value: 7.933531916622121
  - type: nauc_precision_at_100_diff1 value: -8.247029644152319
  - type: nauc_precision_at_100_max value: 10.86740140944616
  - type: nauc_precision_at_100_std value: 9.581885544675918
  - type: nauc_precision_at_10_diff1 value: -2.409043695429943
  - type: nauc_precision_at_10_max value: 21.04733206074314
  - type: nauc_precision_at_10_std value: 10.03334651647101
  - type: nauc_precision_at_1_diff1 value: 62.23623985611396
  - type: nauc_precision_at_1_max value: 45.04945770947091
  - type: nauc_precision_at_1_std value: -0.8804967656575725
  - type: nauc_precision_at_20_diff1 value: -5.230303656931621
  - type: nauc_precision_at_20_max value: 17.77799716919181
  - type: nauc_precision_at_20_std value: 10.739127998618654
  - type: nauc_precision_at_3_diff1 value: 10.40376424999862
  - type: nauc_precision_at_3_max value: 30.933333400254035
  - type: nauc_precision_at_3_std value: 6.126209127968004
  - type: nauc_precision_at_5_diff1 value: 3.147398101830739
  - type: nauc_precision_at_5_max value: 27.1746309955971
  - type: nauc_precision_at_5_std value: 8.874723615388788
  - type: nauc_recall_at_1000_diff1 value: 5.055940692380908
  - type: nauc_recall_at_1000_max value: 22.42031123370267
  - type: nauc_recall_at_1000_std value: 27.75476692527869
  - type: nauc_recall_at_100_diff1 value: 17.86391178198642
  - type: nauc_recall_at_100_max value: 34.776134863678955
  - type: nauc_recall_at_100_std value: 18.96377158778504
  - type: nauc_recall_at_10_diff1 value: 24.863097695413597
  - type: nauc_recall_at_10_max value: 37.697411651507444
  - type: nauc_recall_at_10_std value: 9.519849994253967
  - type: nauc_recall_at_1_diff1 value: 49.56605205141156
  - type: nauc_recall_at_1_max value: 26.251096698710384
  - type: nauc_recall_at_1_std value: -4.580748485387834
  - type: nauc_recall_at_20_diff1 value: 22.440602811005636
  - type: nauc_recall_at_20_max value: 39.538861316515
  - type: nauc_recall_at_20_std value: 11.363269553121468
  - type: nauc_recall_at_3_diff1 value: 32.80302839873736
  - type: nauc_recall_at_3_max value: 32.53105685012729
  - type: nauc_recall_at_3_std value: -0.7140166410605693
  - type: nauc_recall_at_5_diff1 value: 29.375386639154865
  - type: nauc_recall_at_5_max value: 36.91045781164083
  - type: nauc_recall_at_5_std value: 4.725419050262578
  - type: ndcg_at_1 value: 67.13900000000001
  - type: ndcg_at_10 value: 72.392
  - type: ndcg_at_100 value: 75.25800000000001
  - type: ndcg_at_1000 value: 75.982
  - type: ndcg_at_20 value: 73.783
  - type: ndcg_at_3 value: 67.269
  - type: ndcg_at_5 value: 69.807
  - type: precision_at_1 value: 67.13900000000001
  - type: precision_at_10 value: 13.327
  - type: precision_at_100 value: 1.5559999999999998
  - type: precision_at_1000 value: 0.164
  - type: precision_at_20 value: 7.119000000000001
  - type: precision_at_3 value: 35.599
  - type: precision_at_5 value: 23.936
  - type: recall_at_1 value: 47.370000000000005
  - type: recall_at_10 value: 82.16
  - type: recall_at_100 value: 93.34
  - type: recall_at_1000 value: 98.202
  - type: recall_at_20 value: 86.687
  - type: recall_at_3 value: 69.319
  - type: recall_at_5 value: 75.637 task: type: Retrieval
- dataset: config: default name: MTEB RuReviewsClassification (default) revision: f6d2c31f4dc6b88f468552750bfec05b4b41b05a split: test type: ai-forever/ru-reviews-classification metrics:
  - type: accuracy value: 75.0537109375
  - type: f1 value: 74.00523205209554
  - type: f1_weighted value: 74.00436782840376
  - type: main_score value: 75.0537109375 task: type: Classification
- dataset: config: default name: MTEB RuSTSBenchmarkSTS (default) revision: 7cf24f325c6da6195df55bef3d86b5e0616f3018 split: test type: ai-forever/ru-stsbenchmark-sts metrics:
  - type: cosine_pearson value: 81.10255413476487
  - type: cosine_spearman value: 81.40020843157141
  - type: euclidean_pearson value: 81.25155479902466
  - type: euclidean_spearman value: 81.40020831064922
  - type: main_score value: 81.40020843157141
  - type: manhattan_pearson value: 81.1493715249014
  - type: manhattan_spearman value: 81.30973667941649
  - type: pearson value: 81.10255413476487
  - type: spearman value: 81.40020843157141 task: type: STS
- dataset: config: default name: MTEB RuSciBenchGRNTIClassification (default) revision: 673a610d6d3dd91a547a0d57ae1b56f37ebbf6a1 split: test type: ai-forever/ru-scibench-grnti-classification metrics:
  - type: accuracy value: 69.8974609375
  - type: f1 value: 68.57837564785511
  - type: f1_weighted value: 68.59030489460784
  - type: main_score value: 69.8974609375 task: type: Classification
- dataset: config: default name: MTEB RuSciBenchGRNTIClusteringP2P (default) revision: 673a610d6d3dd91a547a0d57ae1b56f37ebbf6a1 split: test type: ai-forever/ru-scibench-grnti-classification metrics:
  - type: main_score value: 67.03880348548029
  - type: v_measure value: 67.03880348548029
  - type: v_measure_std value: 0.6126278133139618 task: type: Clustering
- dataset: config: default name: MTEB RuSciBenchOECDClassification (default) revision: 26c88e99dcaba32bb45d0e1bfc21902337f6d471 split: test type: ai-forever/ru-scibench-oecd-classification metrics:
  - type: accuracy value: 54.63378906250001
  - type: f1 value: 51.34306420274629
  - type: f1_weighted value: 51.33495867493914
  - type: main_score value: 54.63378906250001 task: type: Classification
- dataset: config: default name: MTEB RuSciBenchOECDClusteringP2P (default) revision: 26c88e99dcaba32bb45d0e1bfc21902337f6d471 split: test type: ai-forever/ru-scibench-oecd-classification metrics:
  - type: main_score value: 56.55947121159027
  - type: v_measure value: 56.55947121159027
  - type: v_measure_std value: 0.5498882006880662 task: type: Clustering
- dataset: config: ru name: MTEB STS22 (ru) revision: de9d86b3b84231dc21f76c7b7af1f28e2f57f6e3 split: test type: mteb/sts22-crosslingual-sts metrics:
  - type: cosine_pearson value: 61.833294921667914
  - type: cosine_spearman value: 63.53967536726357
  - type: euclidean_pearson value: 60.382865218855805
  - type: euclidean_spearman value: 63.53967536726357
  - type: main_score value: 63.53967536726357
  - type: manhattan_pearson value: 60.24879015304578
  - type: manhattan_spearman value: 63.42305760430092
  - type: pearson value: 61.833294921667914
  - type: spearman value: 63.53967536726357 task: type: STS
- dataset: config: default name: MTEB SensitiveTopicsClassification (default) revision: 416b34a802308eac30e4192afc0ff99bb8dcc7f2 split: test type: ai-forever/sensitive-topics-classification metrics:
  - type: accuracy value: 39.8193359375
  - type: f1 value: 55.46591740935434
  - type: lrap value: 66.50980631510454
  - type: main_score value: 39.8193359375 task: type: MultilabelClassification
- dataset: config: default name: MTEB TERRa (default) revision: 7b58f24536063837d644aab9a023c62199b2a612 split: dev type: ai-forever/terra-pairclassification metrics:
  - type: cosine_accuracy value: 66.77524429967427
  - type: cosine_accuracy_threshold value: 55.58975338935852
  - type: cosine_ap value: 66.4567219323658
  - type: cosine_f1 value: 70.64676616915423
  - type: cosine_f1_threshold value: 45.55969536304474
  - type: cosine_precision value: 57.028112449799195
  - type: cosine_recall value: 92.81045751633987
  - type: dot_accuracy value: 66.77524429967427
  - type: dot_accuracy_threshold value: 55.589759349823
  - type: dot_ap value: 66.4567219323658
  - type: dot_f1 value: 70.64676616915423
  - type: dot_f1_threshold value: 45.55969536304474
  - type: dot_precision value: 57.028112449799195
  - type: dot_recall value: 92.81045751633987
  - type: euclidean_accuracy value: 66.77524429967427
  - type: euclidean_accuracy_threshold value: 94.24455165863037
  - type: euclidean_ap value: 66.4567219323658
  - type: euclidean_f1 value: 70.64676616915423
  - type: euclidean_f1_threshold value: 104.34587001800537
  - type: euclidean_precision value: 57.028112449799195
  - type: euclidean_recall value: 92.81045751633987
  - type: main_score value: 66.4567219323658
  - type: manhattan_accuracy value: 66.77524429967427
  - type: manhattan_accuracy_threshold value: 2865.5345916748047
  - type: manhattan_ap value: 66.26659863769075
  - type: manhattan_f1 value: 70.8542713567839
  - type: manhattan_f1_threshold value: 3212.3912811279297
  - type: manhattan_precision value: 57.55102040816327
  - type: manhattan_recall value: 92.15686274509804
  - type: max_accuracy value: 66.77524429967427
  - type: max_ap value: 66.4567219323658
  - type: max_f1 value: 70.8542713567839
  - type: max_precision value: 57.55102040816327
  - type: max_recall value: 92.81045751633987
  - type: similarity_accuracy value: 66.77524429967427
  - type: similarity_accuracy_threshold value: 55.58975338935852
  - type: similarity_ap value: 66.4567219323658
  - type: similarity_f1 value: 70.64676616915423
  - type: similarity_f1_threshold value: 45.55969536304474
  - type: similarity_precision value: 57.028112449799195
  - type: similarity_recall value: 92.81045751633987 task: type: PairClassification license: mit language:
ru
en tags:
mteb
transformers
sentence-transformers base_model: ai-forever/FRED-T5-1.7B pipeline_tag: feature-extraction

Model Card for FRIDA

FRIDA is a full-scale finetuned general text embedding model inspired by denoising architecture based on T5. The model is based on the encoder part of FRED-T5 model and continues research of text embedding models (ruMTEB, ru-en-RoSBERTa). It has been pre-trained on a Russian-English dataset and fine-tuned for improved performance on the target task.

For more model details please refer to our technical report [TODO].

Usage

The model can be used as is with prefixes. It is recommended to use CLS pooling. The choice of prefix and pooling depends on the task.

We use the following basic rules to choose a prefix:

"search_query: " and "search_document: " prefixes are for answer or relevant paragraph retrieval
"paraphrase: " prefix is for symmetric paraphrasing related tasks (STS, paraphrase mining, deduplication)
"categorize: " prefix is for asymmetric matching of document title and body (e.g. news, scientific papers, social posts)
"categorize_sentiment: " prefix is for any tasks that rely on sentiment features (e.g. hate, toxic, emotion)
"categorize_topic: " prefix is intended for tasks where you need to group texts by topic
"categorize_entailment: " prefix is for textual entailment task (NLI)

To better tailor the model to your needs, you can fine-tune it with relevant high-quality Russian and English datasets.

Below are examples of texts encoding using the Transformers and SentenceTransformers libraries.

Transformers

import torch
import torch.nn.functional as F
from transformers import AutoTokenizer, T5EncoderModel


def pool(hidden_state, mask, pooling_method="cls"):
    if pooling_method == "mean":
        s = torch.sum(hidden_state * mask.unsqueeze(-1).float(), dim=1)
        d = mask.sum(axis=1, keepdim=True).float()
        return s / d
    elif pooling_method == "cls":
        return hidden_state[:, 0]

inputs = [
    # 
    "paraphrase: В Ярославской области разрешили работу бань, но без посетителей",
    "categorize_entailment: Женщину доставили в больницу, за ее жизнь сейчас борются врачи.",
    "search_query: Сколько программистов нужно, чтобы вкрутить лампочку?",
    # 
    "paraphrase: Ярославским баням разрешили работать без посетителей",
    "categorize_entailment: Женщину спасают врачи.",
    "search_document: Чтобы вкрутить лампочку, требуется три программиста: один напишет программу извлечения лампочки, другой — вкручивания лампочки, а третий проведет тестирование."
]

tokenizer = AutoTokenizer.from_pretrained("ai-forever/FRIDA")
model = T5EncoderModel.from_pretrained("ai-forever/FRIDA")

tokenized_inputs = tokenizer(inputs, max_length=512, padding=True, truncation=True, return_tensors="pt")

with torch.no_grad():
    outputs = model(**tokenized_inputs)
    
embeddings = pool(
    outputs.last_hidden_state, 
    tokenized_inputs["attention_mask"],
    pooling_method="cls" # or try "mean"
)

embeddings = F.normalize(embeddings, p=2, dim=1)
sim_scores = embeddings[:3] @ embeddings[3:].T
print(sim_scores.diag().tolist())
# [0.9360030293464661, 0.8591322302818298, 0.728583037853241]

SentenceTransformers

from sentence_transformers import SentenceTransformer

inputs = [
    # 
    "paraphrase: В Ярославской области разрешили работу бань, но без посетителей",
    "categorize_entailment: Женщину доставили в больницу, за ее жизнь сейчас борются врачи.",
    "search_query: Сколько программистов нужно, чтобы вкрутить лампочку?",
    # 
    "paraphrase: Ярославским баням разрешили работать без посетителей",
    "categorize_entailment: Женщину спасают врачи.",
    "search_document: Чтобы вкрутить лампочку, требуется три программиста: один напишет программу извлечения лампочки, другой — вкручивания лампочки, а третий проведет тестирование."
]

# loads model with CLS pooling
model = SentenceTransformer("ai-forever/FRIDA")

# embeddings are normalized by default
embeddings = model.encode(inputs, convert_to_tensor=True)

sim_scores = embeddings[:3] @ embeddings[3:].T
print(sim_scores.diag().tolist())
# [0.9360026717185974, 0.8591331243515015, 0.7285830974578857]

or using prompts (sentence-transformers>=2.4.0):

from sentence_transformers import SentenceTransformer

# loads model with CLS pooling
model = SentenceTransformer("ai-forever/FRIDA")

paraphrase = model.encode(["В Ярославской области разрешили работу бань, но без посетителей", "Ярославским баням разрешили работать без посетителей"], prompt_name="paraphrase")
print(paraphrase[0] @ paraphrase[1].T) # 0.9360032

categorize_entailment = model.encode(["Женщину доставили в больницу, за ее жизнь сейчас борются врачи.", "Женщину спасают врачи."], prompt_name="categorize_entailment")
print(categorize_entailment[0] @ categorize_entailment[1].T) # 0.8591322

query_embedding = model.encode("Сколько программистов нужно, чтобы вкрутить лампочку?", prompt_name="search_query")
document_embedding = model.encode("Чтобы вкрутить лампочку, требуется три программиста: один напишет программу извлечения лампочки, другой — вкручивания лампочки, а третий проведет тестирование.", prompt_name="search_document")
print(query_embedding @ document_embedding.T) # 0.7285831