List your company
ResearchFormerly WizWand

Conversational Question Answering on CoQA (dev)

0.849Overall F1UNILM
0.252040.407020.5620.71698Sep 27, 2018Dec 17, 2018Mar 8, 2019May 28, 2019Aug 17, 2019Nov 6, 2019Jan 26, 2020
Updated 8mo ago

Evaluation results

MethodLinks
UNILMMode=fine-tuned, Epochs=2, Batch size=16, Maximum length=5122019.050.849
ERNIE-GENLARGEbeam size=32020.010.845
BERT_LARGEModel type=cased, Mode=fine-tuned, Epochs=2, Batch size=16, Maximum length=5122019.050.827
UNILMLARGE2020.010.825
UNILMfine-tuning epochs=10, batch size=32, mask probability=0.5, maximum length=512, label smoothing rate=0.1, beam size=32019.050.825
BiDAF++Context window=3-ctx2018.090.692
DrQA+ELMoArchitecture=LSTM-based, Augmentation=pre-trained ELMo representation2019.050.672
DrQA + PGNetType=Abstractive2018.090.662
BiDAF++Context window=0-ctx2018.090.634
DrQAType=Extractive2018.090.547
PGNet2020.010.454
PGNetarchitecture=Seq2Seq with a copy mechanism2019.050.454
Seq2Seq2020.010.275
Seq2Seqarchitecture=sequence-to-sequence model with an attention mechanism2019.050.275