Semantic Textual Similarity On Sts13

평가 지표

Spearman Correlation

평가 결과

이 벤치마크에서 각 모델의 성능 결과

모델 이름	Spearman Correlation	Paper Title	Repository
SimCSE-RoBERTa-base	0.8136	SimCSE: Simple Contrastive Learning of Sentence Embeddings
DiffCSE-BERT-base	0.8443	DiffCSE: Difference-based Contrastive Learning for Sentence Embeddings
Trans-Encoder-BERT-base-cross (unsup.)	0.8559	Trans-Encoder: Unsupervised sentence-pair modelling through self- and mutual-distillations
PromptEOL+CSE+LLaMA-30B	0.9025	Scaling Sentence Embeddings with Large Language Models
Trans-Encoder-RoBERTa-large-cross (unsup.)	0.8831	Trans-Encoder: Unsupervised sentence-pair modelling through self- and mutual-distillations
SBERT-NLI-large	0.7846	Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks
PromptEOL+CSE+OPT-13B	0.9024	Scaling Sentence Embeddings with Large Language Models
Trans-Encoder-BERT-large-bi (unsup.)	0.8851	Trans-Encoder: Unsupervised sentence-pair modelling through self- and mutual-distillations
PromCSE-RoBERTa-large (0.355B)	0.8897	Improved Universal Sentence Embeddings with Prompt-based Contrastive Learning and Energy-based Learning
AnglE-LLaMA-7B-v2	0.9056	AnglE-optimized Text Embeddings
SimCSE-RoBERTa-large	0.8727	SimCSE: Simple Contrastive Learning of Sentence Embeddings
BERTlarge-flow (target)	0.7339	On the Sentence Embeddings from Pre-trained Language Models	-
Mirror-RoBERTa-base (unsup.)	0.819	Fast, Effective, and Self-Supervised: Transforming Masked Language Models into Universal Lexical and Sentence Encoders
Dino (STSb/̄	0.8126	Generating Datasets with Pretrained Language Models
DiffCSE-RoBERTa-base	0.8343	DiffCSE: Difference-based Contrastive Learning for Sentence Embeddings
AnglE-LLaMA-7B	0.9058	AnglE-optimized Text Embeddings
Trans-Encoder-BERT-large-cross (unsup.)	0.8831	Trans-Encoder: Unsupervised sentence-pair modelling through self- and mutual-distillations
Mirror-BERT-base (unsup.)	0.796	Fast, Effective, and Self-Supervised: Transforming Masked Language Models into Universal Lexical and Sentence Encoders
SimCSE-BERT-base	0.8241	SimCSE: Simple Contrastive Learning of Sentence Embeddings
Trans-Encoder-BERT-base-bi (unsup.)	0.851	Trans-Encoder: Unsupervised sentence-pair modelling through self- and mutual-distillations

0 of 22 row(s) selected.