List your company
ResearchFormerly WizWand

Conditional Semantic Textual Similarity on C-STS (official)

0.4818Spearman CorrelationPoLi-RL (Qwen3-8B)
0.1398480.2286240.31740.406176Oct 5, 2025
Updated 8mo ago

Evaluation results

MethodLinks
PoLi-RL (Qwen3-8B)Training Paradigm=RL, Parameters=8B2025.100.48180.4827
PoLi-RL (Qwen3-4B)Training Paradigm=RL, Parameters=4B2025.100.46230.4619
PoLi-RL (Qwen3-0.6B)Training Paradigm=RL, Parameters=0.6B2025.100.44340.4436
GPT-4oTraining Paradigm=Few-shot2025.100.44230.4407
SEAVER SimCSELARGETraining Paradigm=SFT, Parameters=355M2025.100.43830.4381
GPT-4Training Paradigm=Few-shot2025.100.436—
SimCSELARGETraining Paradigm=SFT, Parameters=355M2025.100.4320.432
DeepSeek-R1Training Paradigm=Few-shot2025.100.42850.4236
RoBERTaLARGETraining Paradigm=SFT, Parameters=355M2025.100.4070.408
Qwen3-8BTraining Paradigm=SFT, Parameters=8B2025.100.40420.4083
Qwen3-4BTraining Paradigm=SFT, Parameters=4B2025.100.38410.3945
Qwen3-4BTraining Paradigm=Few-shot, Parameters=4B2025.100.37970.3848
Qwen3-8BTraining Paradigm=Few-shot, Parameters=8B2025.100.3790.3854
Qwen3-0.6BTraining Paradigm=SFT, Parameters=0.6B2025.100.35590.3683
Flan-T5XXLTraining Paradigm=Few-shot, Parameters=11B2025.100.306—
Qwen3-0.6BTraining Paradigm=Few-shot, Parameters=0.6B2025.100.25250.2519
Flan-UL2Training Paradigm=Few-shot, Parameters=20B2025.100.235—
Tk-InstructTraining Paradigm=Few-shot, Parameters=11B2025.100.178—
GPT-3.5Training Paradigm=Few-shot, Parameters=175B2025.100.153—