Natural Language Inference On Multinli

评估指标

Matched

Mismatched

评测结果

各个模型在此基准测试上的表现结果

			Paper Title	Repository
UnitedSynT5 (3B)	92.6	-	First Train to Generate, then Generate to Train: UnitedSynT5 for Few-Shot NLI	-
Turing NLR v5 XXL 5.4B (fine-tuned)	92.6	92.4	-	-
T5-XXL 11B (fine-tuned)	92.0	-	Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer
T5	92.0	91.7	SMART: Robust and Efficient Fine-Tuning for Pre-trained Natural Language Models through Principled Regularized Optimization
T5-3B	91.4	91.2	Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer
ALBERT	91.3	-	ALBERT: A Lite BERT for Self-supervised Learning of Language Representations
Adv-RoBERTa ensemble	91.1	90.7	StructBERT: Incorporating Language Structures into Pre-training for Deep Language Understanding	-
DeBERTa (large)	91.1	91.1	DeBERTa: Decoding-enhanced BERT with Disentangled Attention
RoBERTa	90.8	-	RoBERTa: A Robustly Optimized BERT Pretraining Approach
XLNet (single model)	90.8	-	XLNet: Generalized Autoregressive Pretraining for Language Understanding
RoBERTa-large 355M (MLP quantized vector-wise, fine-tuned)	90.2	-	LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale
T5-Large	89.9	-	Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer
PSQ (Chen et al., 2020)	89.9	-	A Statistical Framework for Low-bitwidth Training of Deep Neural Networks
UnitedSynT5 (335M)	89.8	-	First Train to Generate, then Generate to Train: UnitedSynT5 for Few-Shot NLI	-
ERNIE 2.0 Large	88.7	88.8	ERNIE 2.0: A Continual Pre-training Framework for Language Understanding
SpanBERT	88.1	-	SpanBERT: Improving Pre-training by Representing and Predicting Spans
ASA + RoBERTa	88	-	Adversarial Self-Attention for Language Understanding
BERT-Large	88	88	FNet: Mixing Tokens with Fourier Transforms
MT-DNN-ensemble	87.9	87.4	Improving Multi-Task Deep Neural Networks via Knowledge Distillation for Natural Language Understanding
Q-BERT (Shen et al., 2020)	87.8	-	Q-BERT: Hessian Based Ultra Low Precision Quantization of BERT	-

0 of 67 row(s) selected.