metadata
tags:
- mteb
- sentence-transformers
model-index:
- name: piccolo-large-zh-v2
results:
- task:
type: STS
dataset:
type: C-MTEB/AFQMC
name: MTEB AFQMC
config: default
split: validation
revision: None
metrics:
- type: cos_sim_pearson
value: 56.76055988260572
- type: cos_sim_spearman
value: 61.49271876861677
- type: euclidean_pearson
value: 59.14524585320711
- type: euclidean_spearman
value: 60.63579339225774
- type: manhattan_pearson
value: 59.14662752965445
- type: manhattan_spearman
value: 60.635190265737904
- task:
type: STS
dataset:
type: C-MTEB/ATEC
name: MTEB ATEC
config: default
split: test
revision: None
metrics:
- type: cos_sim_pearson
value: 56.21706298831197
- type: cos_sim_spearman
value: 59.19831457688953
- type: euclidean_pearson
value: 62.37752017633299
- type: euclidean_spearman
value: 58.79400967473204
- type: manhattan_pearson
value: 62.37015943212308
- type: manhattan_spearman
value: 58.79232537600814
- task:
type: Classification
dataset:
type: mteb/amazon_reviews_multi
name: MTEB AmazonReviewsClassification (zh)
config: zh
split: test
revision: 1399c76144fd37290681b995c656ef9b2e06e26d
metrics:
- type: accuracy
value: 49.440000000000005
- type: f1
value: 46.67381446305019
- task:
type: STS
dataset:
type: C-MTEB/BQ
name: MTEB BQ
config: default
split: test
revision: None
metrics:
- type: cos_sim_pearson
value: 70.99026329599994
- type: cos_sim_spearman
value: 72.87565357908989
- type: euclidean_pearson
value: 71.17690439270028
- type: euclidean_spearman
value: 72.50428109969029
- type: manhattan_pearson
value: 71.17262321033088
- type: manhattan_spearman
value: 72.49845447987437
- task:
type: Clustering
dataset:
type: C-MTEB/CLSClusteringP2P
name: MTEB CLSClusteringP2P
config: default
split: test
revision: None
metrics:
- type: v_measure
value: 57.92713421071616
- task:
type: Clustering
dataset:
type: C-MTEB/CLSClusteringS2S
name: MTEB CLSClusteringS2S
config: default
split: test
revision: None
metrics:
- type: v_measure
value: 48.096546680932235
- task:
type: Reranking
dataset:
type: C-MTEB/CMedQAv1-reranking
name: MTEB CMedQAv1
config: default
split: test
revision: None
metrics:
- type: map
value: 89.31003741715936
- type: mrr
value: 91.38075396825397
- task:
type: Reranking
dataset:
type: C-MTEB/CMedQAv2-reranking
name: MTEB CMedQAv2
config: default
split: test
revision: None
metrics:
- type: map
value: 90.13769781784876
- type: mrr
value: 92.14329365079365
- task:
type: Retrieval
dataset:
type: C-MTEB/CmedqaRetrieval
name: MTEB CmedqaRetrieval
config: default
split: dev
revision: None
metrics:
- type: map_at_1
value: 26.931
- type: map_at_10
value: 40.647
- type: map_at_100
value: 42.519
- type: map_at_1000
value: 42.616
- type: map_at_3
value: 36.144999999999996
- type: map_at_5
value: 38.717
- type: mrr_at_1
value: 40.935
- type: mrr_at_10
value: 49.684
- type: mrr_at_100
value: 50.598
- type: mrr_at_1000
value: 50.632999999999996
- type: mrr_at_3
value: 47.07
- type: mrr_at_5
value: 48.49
- type: ndcg_at_1
value: 40.935
- type: ndcg_at_10
value: 47.583999999999996
- type: ndcg_at_100
value: 54.69199999999999
- type: ndcg_at_1000
value: 56.314
- type: ndcg_at_3
value: 41.973
- type: ndcg_at_5
value: 44.334
- type: precision_at_1
value: 40.935
- type: precision_at_10
value: 10.585
- type: precision_at_100
value: 1.637
- type: precision_at_1000
value: 0.184
- type: precision_at_3
value: 23.881
- type: precision_at_5
value: 17.399
- type: recall_at_1
value: 26.931
- type: recall_at_10
value: 59.006
- type: recall_at_100
value: 88.247
- type: recall_at_1000
value: 99.045
- type: recall_at_3
value: 42.064
- type: recall_at_5
value: 49.266
- task:
type: PairClassification
dataset:
type: C-MTEB/CMNLI
name: MTEB Cmnli
config: default
split: validation
revision: None
metrics:
- type: cos_sim_accuracy
value: 86.08538785327721
- type: cos_sim_ap
value: 92.64373114205229
- type: cos_sim_f1
value: 86.89951395953432
- type: cos_sim_precision
value: 84.11378555798687
- type: cos_sim_recall
value: 89.87608136544307
- type: dot_accuracy
value: 72.66386049308478
- type: dot_ap
value: 81.053422935767
- type: dot_f1
value: 75.19933726830277
- type: dot_precision
value: 67.4907063197026
- type: dot_recall
value: 84.89595510872107
- type: euclidean_accuracy
value: 85.52014431749849
- type: euclidean_ap
value: 91.90647782899615
- type: euclidean_f1
value: 86.26361413647477
- type: euclidean_precision
value: 82.2071595001059
- type: euclidean_recall
value: 90.74117371989713
- type: manhattan_accuracy
value: 85.48406494287433
- type: manhattan_ap
value: 91.89657919524385
- type: manhattan_f1
value: 86.20413761572752
- type: manhattan_precision
value: 84.324686940966
- type: manhattan_recall
value: 88.16927753097966
- type: max_accuracy
value: 86.08538785327721
- type: max_ap
value: 92.64373114205229
- type: max_f1
value: 86.89951395953432
- task:
type: Retrieval
dataset:
type: C-MTEB/CovidRetrieval
name: MTEB CovidRetrieval
config: default
split: dev
revision: None
metrics:
- type: map_at_1
value: 75.50099999999999
- type: map_at_10
value: 83.43
- type: map_at_100
value: 83.577
- type: map_at_1000
value: 83.57900000000001
- type: map_at_3
value: 82.06400000000001
- type: map_at_5
value: 82.88600000000001
- type: mrr_at_1
value: 75.869
- type: mrr_at_10
value: 83.536
- type: mrr_at_100
value: 83.682
- type: mrr_at_1000
value: 83.68299999999999
- type: mrr_at_3
value: 82.244
- type: mrr_at_5
value: 82.998
- type: ndcg_at_1
value: 75.764
- type: ndcg_at_10
value: 86.777
- type: ndcg_at_100
value: 87.36
- type: ndcg_at_1000
value: 87.424
- type: ndcg_at_3
value: 84.10300000000001
- type: ndcg_at_5
value: 85.532
- type: precision_at_1
value: 75.764
- type: precision_at_10
value: 9.8
- type: precision_at_100
value: 1.005
- type: precision_at_1000
value: 0.101
- type: precision_at_3
value: 30.207
- type: precision_at_5
value: 18.82
- type: recall_at_1
value: 75.50099999999999
- type: recall_at_10
value: 96.997
- type: recall_at_100
value: 99.473
- type: recall_at_1000
value: 100
- type: recall_at_3
value: 89.831
- type: recall_at_5
value: 93.256
- task:
type: Retrieval
dataset:
type: C-MTEB/DuRetrieval
name: MTEB DuRetrieval
config: default
split: dev
revision: None
metrics:
- type: map_at_1
value: 27.094
- type: map_at_10
value: 82.418
- type: map_at_100
value: 85.05
- type: map_at_1000
value: 85.083
- type: map_at_3
value: 57.68600000000001
- type: map_at_5
value: 72.476
- type: mrr_at_1
value: 92.25
- type: mrr_at_10
value: 94.621
- type: mrr_at_100
value: 94.675
- type: mrr_at_1000
value: 94.677
- type: mrr_at_3
value: 94.375
- type: mrr_at_5
value: 94.52199999999999
- type: ndcg_at_1
value: 92.25
- type: ndcg_at_10
value: 89.13600000000001
- type: ndcg_at_100
value: 91.532
- type: ndcg_at_1000
value: 91.836
- type: ndcg_at_3
value: 88.50099999999999
- type: ndcg_at_5
value: 87.251
- type: precision_at_1
value: 92.25
- type: precision_at_10
value: 42.295
- type: precision_at_100
value: 4.812
- type: precision_at_1000
value: 0.48900000000000005
- type: precision_at_3
value: 79.167
- type: precision_at_5
value: 66.56
- type: recall_at_1
value: 27.094
- type: recall_at_10
value: 89.816
- type: recall_at_100
value: 97.855
- type: recall_at_1000
value: 99.384
- type: recall_at_3
value: 59.557
- type: recall_at_5
value: 76.395
- task:
type: Retrieval
dataset:
type: C-MTEB/EcomRetrieval
name: MTEB EcomRetrieval
config: default
split: dev
revision: None
metrics:
- type: map_at_1
value: 53.6
- type: map_at_10
value: 62.985
- type: map_at_100
value: 63.532999999999994
- type: map_at_1000
value: 63.546
- type: map_at_3
value: 60.617
- type: map_at_5
value: 62.017
- type: mrr_at_1
value: 53.6
- type: mrr_at_10
value: 62.985
- type: mrr_at_100
value: 63.532999999999994
- type: mrr_at_1000
value: 63.546
- type: mrr_at_3
value: 60.617
- type: mrr_at_5
value: 62.017
- type: ndcg_at_1
value: 53.6
- type: ndcg_at_10
value: 67.755
- type: ndcg_at_100
value: 70.366
- type: ndcg_at_1000
value: 70.696
- type: ndcg_at_3
value: 62.89900000000001
- type: ndcg_at_5
value: 65.437
- type: precision_at_1
value: 53.6
- type: precision_at_10
value: 8.28
- type: precision_at_100
value: 0.9490000000000001
- type: precision_at_1000
value: 0.098
- type: precision_at_3
value: 23.166999999999998
- type: precision_at_5
value: 15.14
- type: recall_at_1
value: 53.6
- type: recall_at_10
value: 82.8
- type: recall_at_100
value: 94.89999999999999
- type: recall_at_1000
value: 97.5
- type: recall_at_3
value: 69.5
- type: recall_at_5
value: 75.7
- task:
type: Classification
dataset:
type: C-MTEB/IFlyTek-classification
name: MTEB IFlyTek
config: default
split: validation
revision: None
metrics:
- type: accuracy
value: 52.104655636783384
- type: f1
value: 41.025743582860514
- task:
type: Classification
dataset:
type: C-MTEB/JDReview-classification
name: MTEB JDReview
config: default
split: test
revision: None
metrics:
- type: accuracy
value: 88.57410881801127
- type: ap
value: 59.49612312498937
- type: f1
value: 83.70595013666741
- task:
type: STS
dataset:
type: C-MTEB/LCQMC
name: MTEB LCQMC
config: default
split: test
revision: None
metrics:
- type: cos_sim_pearson
value: 74.00327736048256
- type: cos_sim_spearman
value: 79.5459672237356
- type: euclidean_pearson
value: 79.18300205389669
- type: euclidean_spearman
value: 79.21872988987533
- type: manhattan_pearson
value: 79.1715470733081
- type: manhattan_spearman
value: 79.20756273498812
- task:
type: Retrieval
dataset:
type: C-MTEB/MMarcoRetrieval
name: MTEB MMarcoRetrieval
config: default
split: dev
revision: None
metrics:
- type: map_at_1
value: 66.94600000000001
- type: map_at_10
value: 75.947
- type: map_at_100
value: 76.268
- type: map_at_1000
value: 76.28
- type: map_at_3
value: 74.13300000000001
- type: map_at_5
value: 75.28399999999999
- type: mrr_at_1
value: 69.241
- type: mrr_at_10
value: 76.532
- type: mrr_at_100
value: 76.816
- type: mrr_at_1000
value: 76.827
- type: mrr_at_3
value: 74.95
- type: mrr_at_5
value: 75.957
- type: ndcg_at_1
value: 69.241
- type: ndcg_at_10
value: 79.54299999999999
- type: ndcg_at_100
value: 80.95
- type: ndcg_at_1000
value: 81.252
- type: ndcg_at_3
value: 76.119
- type: ndcg_at_5
value: 78.069
- type: precision_at_1
value: 69.241
- type: precision_at_10
value: 9.576
- type: precision_at_100
value: 1.026
- type: precision_at_1000
value: 0.105
- type: precision_at_3
value: 28.571999999999996
- type: precision_at_5
value: 18.181
- type: recall_at_1
value: 66.94600000000001
- type: recall_at_10
value: 90.024
- type: recall_at_100
value: 96.3
- type: recall_at_1000
value: 98.656
- type: recall_at_3
value: 81.026
- type: recall_at_5
value: 85.658
- 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: 77.71015467383997
- type: f1
value: 74.32345894845358
- 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: 85.63214525891055
- type: f1
value: 84.65303466003252
- task:
type: Retrieval
dataset:
type: C-MTEB/MedicalRetrieval
name: MTEB MedicalRetrieval
config: default
split: dev
revision: None
metrics:
- type: map_at_1
value: 55.50000000000001
- type: map_at_10
value: 61.66199999999999
- type: map_at_100
value: 62.13999999999999
- type: map_at_1000
value: 62.187000000000005
- type: map_at_3
value: 59.967000000000006
- type: map_at_5
value: 60.927
- type: mrr_at_1
value: 55.7
- type: mrr_at_10
value: 61.76199999999999
- type: mrr_at_100
value: 62.241
- type: mrr_at_1000
value: 62.287000000000006
- type: mrr_at_3
value: 60.06700000000001
- type: mrr_at_5
value: 61.027
- type: ndcg_at_1
value: 55.50000000000001
- type: ndcg_at_10
value: 64.878
- type: ndcg_at_100
value: 67.464
- type: ndcg_at_1000
value: 68.745
- type: ndcg_at_3
value: 61.367000000000004
- type: ndcg_at_5
value: 63.117999999999995
- type: precision_at_1
value: 55.50000000000001
- type: precision_at_10
value: 7.51
- type: precision_at_100
value: 0.878
- type: precision_at_1000
value: 0.098
- type: precision_at_3
value: 21.8
- type: precision_at_5
value: 13.94
- type: recall_at_1
value: 55.50000000000001
- type: recall_at_10
value: 75.1
- type: recall_at_100
value: 87.8
- type: recall_at_1000
value: 97.89999999999999
- type: recall_at_3
value: 65.4
- type: recall_at_5
value: 69.69999999999999
- task:
type: Reranking
dataset:
type: C-MTEB/Mmarco-reranking
name: MTEB MMarcoReranking
config: default
split: dev
revision: None
metrics:
- type: map
value: 33.386980266936106
- type: mrr
value: 32.11904761904762
- task:
type: Classification
dataset:
type: C-MTEB/MultilingualSentiment-classification
name: MTEB MultilingualSentiment
config: default
split: validation
revision: None
metrics:
- type: accuracy
value: 79.08666666666666
- type: f1
value: 78.93142205976953
- task:
type: PairClassification
dataset:
type: C-MTEB/OCNLI
name: MTEB Ocnli
config: default
split: validation
revision: None
metrics:
- type: cos_sim_accuracy
value: 84.35300487276665
- type: cos_sim_ap
value: 87.83572265803564
- type: cos_sim_f1
value: 85.42713567839195
- type: cos_sim_precision
value: 81.49568552253116
- type: cos_sim_recall
value: 89.7571277719113
- type: dot_accuracy
value: 72.87493232268544
- type: dot_ap
value: 80.29032993894747
- type: dot_f1
value: 76.5938475256353
- type: dot_precision
value: 66.28086419753086
- type: dot_recall
value: 90.70749736008447
- type: euclidean_accuracy
value: 82.34975636166757
- type: euclidean_ap
value: 85.73873757468064
- type: euclidean_f1
value: 83.56713426853707
- type: euclidean_precision
value: 79.50428979980934
- type: euclidean_recall
value: 88.0675818373812
- type: manhattan_accuracy
value: 82.45804006497022
- type: manhattan_ap
value: 85.7176464290469
- type: manhattan_f1
value: 83.65095285857572
- type: manhattan_precision
value: 79.65616045845272
- type: manhattan_recall
value: 88.0675818373812
- type: max_accuracy
value: 84.35300487276665
- type: max_ap
value: 87.83572265803564
- type: max_f1
value: 85.42713567839195
- task:
type: Classification
dataset:
type: C-MTEB/OnlineShopping-classification
name: MTEB OnlineShopping
config: default
split: test
revision: None
metrics:
- type: accuracy
value: 94.61999999999999
- type: ap
value: 92.74140430219491
- type: f1
value: 94.60775857122515
- task:
type: STS
dataset:
type: C-MTEB/PAWSX
name: MTEB PAWSX
config: default
split: test
revision: None
metrics:
- type: cos_sim_pearson
value: 39.75749234575995
- type: cos_sim_spearman
value: 46.48035295363829
- type: euclidean_pearson
value: 45.38711981599582
- type: euclidean_spearman
value: 46.13915356562481
- type: manhattan_pearson
value: 45.420770530489065
- type: manhattan_spearman
value: 46.179913441143775
- task:
type: STS
dataset:
type: C-MTEB/QBQTC
name: MTEB QBQTC
config: default
split: test
revision: None
metrics:
- type: cos_sim_pearson
value: 44.02008249965321
- type: cos_sim_spearman
value: 45.906917552219156
- type: euclidean_pearson
value: 36.600317631983316
- type: euclidean_spearman
value: 41.97740958824762
- type: manhattan_pearson
value: 36.54329048509785
- type: manhattan_spearman
value: 41.91222171040451
- task:
type: STS
dataset:
type: mteb/sts22-crosslingual-sts
name: MTEB STS22 (zh)
config: zh
split: test
revision: None
metrics:
- type: cos_sim_pearson
value: 60.97044608578288
- type: cos_sim_spearman
value: 63.76187490245927
- type: euclidean_pearson
value: 60.74245987426317
- type: euclidean_spearman
value: 63.32990713078846
- type: manhattan_pearson
value: 60.62422616577702
- type: manhattan_spearman
value: 63.256612476686826
- task:
type: STS
dataset:
type: C-MTEB/STSB
name: MTEB STSB
config: default
split: test
revision: None
metrics:
- type: cos_sim_pearson
value: 76.28185867362305
- type: cos_sim_spearman
value: 78.71478656159289
- type: euclidean_pearson
value: 79.80734359535234
- type: euclidean_spearman
value: 79.85403491297063
- type: manhattan_pearson
value: 79.79454037962215
- type: manhattan_spearman
value: 79.82796402623201
- task:
type: Reranking
dataset:
type: C-MTEB/T2Reranking
name: MTEB T2Reranking
config: default
split: dev
revision: None
metrics:
- type: map
value: 67.14759526113295
- type: mrr
value: 77.36422096484723
- task:
type: Retrieval
dataset:
type: C-MTEB/T2Retrieval
name: MTEB T2Retrieval
config: default
split: dev
revision: None
metrics:
- type: map_at_1
value: 28.177999999999997
- type: map_at_10
value: 78.77199999999999
- type: map_at_100
value: 82.365
- type: map_at_1000
value: 82.422
- type: map_at_3
value: 55.452999999999996
- type: map_at_5
value: 68.12700000000001
- type: mrr_at_1
value: 91.097
- type: mrr_at_10
value: 93.52000000000001
- type: mrr_at_100
value: 93.587
- type: mrr_at_1000
value: 93.589
- type: mrr_at_3
value: 93.136
- type: mrr_at_5
value: 93.381
- type: ndcg_at_1
value: 91.097
- type: ndcg_at_10
value: 86.136
- type: ndcg_at_100
value: 89.515
- type: ndcg_at_1000
value: 90.049
- type: ndcg_at_3
value: 87.41600000000001
- type: ndcg_at_5
value: 86.115
- type: precision_at_1
value: 91.097
- type: precision_at_10
value: 42.597
- type: precision_at_100
value: 5.043
- type: precision_at_1000
value: 0.517
- type: precision_at_3
value: 76.239
- type: precision_at_5
value: 63.93
- type: recall_at_1
value: 28.177999999999997
- type: recall_at_10
value: 85.182
- type: recall_at_100
value: 96.174
- type: recall_at_1000
value: 98.848
- type: recall_at_3
value: 57.150999999999996
- type: recall_at_5
value: 71.50999999999999
- task:
type: Classification
dataset:
type: C-MTEB/TNews-classification
name: MTEB TNews
config: default
split: validation
revision: None
metrics:
- type: accuracy
value: 54.521
- type: f1
value: 52.53528052282081
- task:
type: Clustering
dataset:
type: C-MTEB/ThuNewsClusteringP2P
name: MTEB ThuNewsClusteringP2P
config: default
split: test
revision: None
metrics:
- type: v_measure
value: 74.2003249023509
- task:
type: Clustering
dataset:
type: C-MTEB/ThuNewsClusteringS2S
name: MTEB ThuNewsClusteringS2S
config: default
split: test
revision: None
metrics:
- type: v_measure
value: 68.4277378629746
- task:
type: Retrieval
dataset:
type: C-MTEB/VideoRetrieval
name: MTEB VideoRetrieval
config: default
split: dev
revision: None
metrics:
- type: map_at_1
value: 58.599999999999994
- type: map_at_10
value: 68.671
- type: map_at_100
value: 69.148
- type: map_at_1000
value: 69.157
- type: map_at_3
value: 66.9
- type: map_at_5
value: 68.045
- type: mrr_at_1
value: 58.599999999999994
- type: mrr_at_10
value: 68.671
- type: mrr_at_100
value: 69.148
- type: mrr_at_1000
value: 69.157
- type: mrr_at_3
value: 66.9
- type: mrr_at_5
value: 68.045
- type: ndcg_at_1
value: 58.599999999999994
- type: ndcg_at_10
value: 73.099
- type: ndcg_at_100
value: 75.33
- type: ndcg_at_1000
value: 75.58500000000001
- type: ndcg_at_3
value: 69.502
- type: ndcg_at_5
value: 71.542
- type: precision_at_1
value: 58.599999999999994
- type: precision_at_10
value: 8.68
- type: precision_at_100
value: 0.97
- type: precision_at_1000
value: 0.099
- type: precision_at_3
value: 25.667
- type: precision_at_5
value: 16.38
- type: recall_at_1
value: 58.599999999999994
- type: recall_at_10
value: 86.8
- type: recall_at_100
value: 97
- type: recall_at_1000
value: 99.1
- type: recall_at_3
value: 77
- type: recall_at_5
value: 81.89999999999999
- task:
type: Classification
dataset:
type: C-MTEB/waimai-classification
name: MTEB Waimai
config: default
split: test
revision: None
metrics:
- type: accuracy
value: 89.58999999999999
- type: ap
value: 75.69899834265364
- type: f1
value: 88.2026184757175
News
[2024-05-16]
Due to certain internal company considerations, we have temporarily removed the model weights.
It will be uploaded again after passing our internal review process.
Please temporarily access this model via API: https://platform.sensenova.cn/doc?path=/chat/Embeddings/Embeddings.md
[2024-05-14]
We have currently release our model weights, training code, and tech report. Discussions are welcome.
For training code, please refer to our github
For training details, please refer to our tech-report
[2024-04-22]
piccolo-large-zh-v2 currently ranks first on the C-MTEB list, leading the previous BERT model by about 1.9 points.
Piccolo-large-zh-v2
piccolo-large-zh-v2 is a Chinese embedding model developed by the general model group from SenseTime Research. This upgraded version of Piccolo aims to prioritize general downstream fine-tuning methods. Piccolo2 primarily leverages an efficient multi-task hybrid loss training approach, effectively harnessing textual data and labels from diverse downstream tasks. In addition, Piccolo2 scales up the embedding dimension and uses MRL training to support more flexible vector dimensions.
💡 Model Hightlights
The main feature of piccolo2 is that it uses a multi-task hybrid loss during training.
For retrieval/sorting tasks, we use the standard InfoNCE with in-batch-negative:
For sts/pair classification tasks, we use cosent loss, which is proved to be better for data with more fine-grained labels(e.g. score values ):
For classification/clustering tasks, by treating text and its semantic labels as positive and negative pairs, we convert the dataset into the format of triples. And then we use InfoNCE to optimize it. However, it’s important to stress that in-batch negatives are no longer used due to the fact that it can easily lead to conflict training targets:
📃 Experiments and Results
Piccolo2 primarily focuses on the downstream general finetune paradigm. Our open source model uses stella-v3.5 as initialization and trained about 2500 steps on 32 GPUS. For more implementation details, please refer to our technical report.
| Model Name | Model Size (GB) | Dimension | Sequence Length | Classification (9) | Clustering (4) | Pair Classification (2) | Reranking (4) | Retrieval (8) | STS (8) | Average (35) |
|---|---|---|---|---|---|---|---|---|---|---|
| piccolo-large-zh-v2 | 1.21 | 1792 | 512 | 74.59 | 62.17 | 90.24 | 70 | 74.36 | 63.5 | 70.95 |
| gte-Qwen1.5-7B-instruct | 26.45 | 32768 | 4096 | 73.35 | 67.08 | 88.52 | 66.38 | 70.62 | 62.32 | 69.56 |
| acge-text-embedding | 1.21 | 1792 | 512 | 72.75 | 58.7 | 87.84 | 67.98 | 72.93 | 62.09 | 69.07 |
🔨 Usage
The piccolo model can be easily accessed in the sentence-transformer package:
# for s2s/s2p dataset, you can use piccolo as below
from sklearn.preprocessing import normalize
from sentence_transformers import SentenceTransformer
sentences = ["数据1", "数据2"]
matryoshka_dim=1792 # support 256, 512, 768, 1024, 1280, 1536, 1792
model = SentenceTransformer('sensenova/piccolo-large-zh-v2')
embeddings_1 = model.encode(sentences, normalize_embeddings=False)
embeddings_2 = model.encode(sentences, normalize_embeddings=False)
embeddings_1 = normalize(embeddings_1[..., :matryoshka_dim], norm="l2", axis=1)
embeddings_2 = normalize(embeddings_2[..., :matryoshka_dim], norm="l2", axis=1)
similarity = embeddings_1 @ embeddings_2.T
🤗 Model List
| Model | Language | Description | prompt |
|---|---|---|---|
| sensenova/piccolo-large-zh-v2 | Chinese | version2: finetuning with multi-task hybrid loss training | None |
| sensenova/piccolo-large-zh | Chinese | version1: pretrain under 400 million chinese text pair | '查询'/'结果' |
| sensenova/piccolo-base-zh | Chinese | version1: pretrain under 400 million chinese text pair | '查询'/'结果' |
Citation
If you find our tech report, models or code helpful, please cite our report or give a star on github or huggingface!
@misc{2405.06932,
Author = {Junqin Huang and Zhongjie Hu and Zihao Jing and Mengya Gao and Yichao Wu},
Title = {Piccolo2: General Text Embedding with Multi-task Hybrid Loss Training},
Year = {2024},
Eprint = {arXiv:2405.06932},
}