Natural Language Inference On Qnli

评估指标

Accuracy

评测结果

各个模型在此基准测试上的表现结果

		Paper Title	Repository
ALICE	99.2%	SMART: Robust and Efficient Fine-Tuning for Pre-trained Natural Language Models through Principled Regularized Optimization
ALBERT	99.2%	ALBERT: A Lite BERT for Self-supervised Learning of Language Representations
StructBERTRoBERTa ensemble	99.2%	StructBERT: Incorporating Language Structures into Pre-training for Deep Language Understanding	-
MT-DNN-SMART	99.2%	SMART: Robust and Efficient Fine-Tuning for Pre-trained Natural Language Models through Principled Regularized Optimization
RoBERTa (ensemble)	98.9%	RoBERTa: A Robustly Optimized BERT Pretraining Approach
T5-11B	96.7%	Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer
T5-3B	96.3%	Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer
DeBERTaV3large	96%	DeBERTaV3: Improving DeBERTa using ELECTRA-Style Pre-Training with Gradient-Disentangled Embedding Sharing
ELECTRA	95.4%	-	-
DeBERTa (large)	95.3%	DeBERTa: Decoding-enhanced BERT with Disentangled Attention
XLNet (single model)	94.9%	XLNet: Generalized Autoregressive Pretraining for Language Understanding
T5-Large 770M	94.8%	Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer
RoBERTa-large 355M (MLP quantized vector-wise, fine-tuned)	94.7%	LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale
ERNIE 2.0 Large	94.6%	ERNIE 2.0: A Continual Pre-training Framework for Language Understanding
PSQ (Chen et al., 2020)	94.5	A Statistical Framework for Low-bitwidth Training of Deep Neural Networks
RoBERTa-large 355M + Entailment as Few-shot Learner	94.5%	Entailment as Few-Shot Learner
SpanBERT	94.3%	SpanBERT: Improving Pre-training by Representing and Predicting Spans
TRANS-BLSTM	94.08%	TRANS-BLSTM: Transformer with Bidirectional LSTM for Language Understanding	-
T5-Base	93.7%	Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer
ASA + RoBERTa	93.6%	Adversarial Self-Attention for Language Understanding

0 of 43 row(s) selected.