List your company
ResearchFormerly WizWand

Natural Language Inference on MultiNLI Mismatched

79.1AccuracyDensely Interactive Inference Network
33.8645.60557.3569.095Dec 30, 2017Nov 19, 2018Oct 9, 2019Aug 28, 2020Jul 18, 2021Jun 7, 2022Apr 27, 2023
Updated 8mo ago

Evaluation results

MethodLinks
Densely Interactive Inference Network2018.1179.1
CAFE Ensembleensemble_size=5 models2017.1279
CAFE2017.1277.9
Compare-Propagate Alignment-Factorized Encoders2018.1177.9
ESIM + Readsource=Weissenborn, 20172017.1277
ESIMsource=Weissenborn, 20172017.1275.8
CDME*Number of parameters=9.0M, Encoder=BiLSTM-Max, Embedding sets=multiple2018.0474.9
Chen et al.2018.1174.9
Unweighted DMENumber of parameters=8.6M, Encoder=BiLSTM-Max2018.0474.4
DMENumber of parameters=8.6M, Encoder=BiLSTM-Max2018.0474.4
DME*Number of parameters=9.0M, Encoder=BiLSTM-Max, Embedding sets=multiple2018.0474.3
CDMENumber of parameters=8.6M, Encoder=BiLSTM-Max2018.0474.1
BiLSTM generalized poolingpenalization=on parameter matrices2018.0674
aESIMimplementation=implemented on Keras2018.1273.9
SSE2018.0473.6
Nie and Bansal2018.1173.6
BiLSTM gated-pooling2018.0673.6
Shortcut stacked BiLSTM2018.0673.6
ESIMimplementation=implemented on Keras2018.1273.5
Naive baselineNumber of parameters=61.3M, Encoder=2048-dimensional BiLSTM2018.0473
BiLSTM max pooling2018.0673
Distance-Based Self-Attention Network2018.1172.9
Stacked Bi-LSTMs + shortcut connections + max-poolinghidden_state_dim=300, bidirectional_dim=600, optimizer=Adam2018.1172.2
ESIMsource=Williams et al., 2017 / Khot et al., 20182017.1272.1
Enhanced Sequential Inference Model2018.1171.9
BiLSTM last pooling2018.0671.9
BiLSTM mean pooling2018.0671.6
Directional Self-Attention Encoders2018.1171.4
Naive baselineNumber of parameters=9.8M, Encoder=512-dimensional BiLSTM2018.0471.1
Bi-LSTM sentence encoder + max-poolinghidden_state_dim=300, bidirectional_dim=600, optimizer=Adam2018.1171.1
Bi-LSTM sentence encoder + max-pooling + attentionhidden_state_dim=300, bidirectional_dim=600, optimizer=Adam2018.1170.8
Stacked Bi-LSTMs + shortcut connections + max-pooling + attentionhidden_state_dim=300, bidirectional_dim=600, optimizer=Adam2018.1170.5
Bi-LSTM sentence encoder + attentionhidden_state_dim=300, bidirectional_dim=600, optimizer=Adam2018.1170.4
FastText BiLSTM-MaxNumber of parameters=8.6M, Embedding type=FastText, Encoder=BiLSTM-Max2018.0470.3
GloVe BiLSTM-MaxNumber of parameters=8.6M, Embedding type=GloVe, Encoder=BiLSTM-Max2018.0470
Bi-GRU sentence encoder + attentionhidden_state_dim=300, bidirectional_dim=600, optimizer=Adam2018.1169.5
BiLSTM2017.1269.4
Stacked Bi-GRUs + shortcut connections + max-poolinghidden_state_dim=300, bidirectional_dim=600, optimizer=Adam2018.1168.9
Stacked Bi-GRUs + shortcut connections + max-pooling + attentionhidden_state_dim=300, bidirectional_dim=600, optimizer=Adam2018.1168.4
Bi-GRU sentence encoder + max-pooling + attentionhidden_state_dim=300, bidirectional_dim=600, optimizer=Adam2018.1168.2
SWEM-max2018.0567.7
SWEM-concat2018.0567.6
Bi-GRU sentence encoder + max-poolinghidden_state_dim=300, bidirectional_dim=600, optimizer=Adam2018.1167.3
BiLSTM2018.1167.1
LSTMType=Bidirectional2018.0566.9
Bi-LSTMsource=baseline from [17]2018.1266.9
BiLSTM2018.0666.9
SWEM-aver2018.0566.2
CNN2018.0565.3
CBOW2017.1264.8
Continuous BOW (Averaging Word Embeddings)2018.1164.6
CBOWsource=baseline from [17]2018.1264.5
CBOW2018.0664.5
LaMini-GPT-J# of params=6B2023.0464
LaMini-LLaMA# of params=7B2023.0463.8
Alpaca# of params=7B2023.0439.6
GPT-J# of params=6B2023.0437.7
Majority2017.1235.6
Most Frequent Class2018.1135.6
LLaMA# of params=7B2023.0435.6