List your company
ResearchFormerly WizWand

Text Classification on SST-2 (Standard Accuracy)

94.8AccuracyICL
53.26474.885.6Mar 29, 2025Jun 7, 2025Aug 16, 2025Oct 25, 2025Jan 3, 2026Mar 14, 2026May 24, 2026
Updated 5mo ago

Evaluation results

MethodLinks
ICLBackbone=LLaMA-2-13B2026.0594.8
ICLBackbone=Qwen-3-8B2026.0594.48
ICLBackbone=Qwen-2.5-7B2026.0594.4
ICLBackbone=LLaMA-3.1-8B2026.0594.2
Task VectorBackbone=Qwen-2.5-7B2026.0593.76
ICLBackbone=LLaMA-2-7B2026.0593.52
ConMeZOBackbone=RoBERTa-Large, Training Steps=10K2025.1193.5
I2CLBackbone=Qwen-3-8B2026.0593.24
Task VectorBackbone=Qwen-3-8B2026.0593.16
BERT-baseParams=110M, Attention=true, Tokenization=true2026.0493
State VectorBackbone=Qwen-3-8B2026.0592.24
Zero-shotBackbone=Qwen-3-8B2026.0592.2
ByT5-SmallParams=300M, Attention=true, Tokenization=false2026.0492
LTVBackbone=Qwen-2.5-7B2026.0591.76
Function VectorBackbone=Qwen-3-8B2026.0591.36
DistilBERTParams=66M, Attention=true, Tokenization=true2026.0491
I2CLBackbone=LLaMA-3.1-8B2026.0590.8
ConMeZOBackbone=OPT-1.3B, Training Steps=20K2025.1190.6
LTVBackbone=Qwen-3-8B2026.0590.4
LTVBackbone=LLaMA-2-7B2026.0590.12
LTVBackbone=LLaMA-3.1-8B2026.0590.08
ZO-AdaMMBackbone=RoBERTa-Large, Training Steps=20K2025.1189.8
LTVBackbone=LLaMA-2-13B2026.0588.96
Task VectorBackbone=LLaMA-3.1-8B2026.0587.88
I2CLBackbone=Qwen-2.5-7B2026.0587.36
Function VectorBackbone=LLaMA-3.1-8B2026.0587.32
State VectorBackbone=LLaMA-2-13B2026.0587.2
Zero-shotBackbone=LLaMA-3.1-8B2026.0586.8
State VectorBackbone=Qwen-2.5-7B2026.0586.64
CANINE-SParams=132M, Attention=true, Tokenization=false2026.0485.8
ZO-AdaMMBackbone=OPT-1.3B, Training Steps=20K2025.1184.4
State VectorBackbone=LLaMA-3.1-8B2026.0584.12
USMoEBackbone=Transformer-XL(20M), Top-k=1.5, FLOPs=6.6753 x 10^10, Fine-tuning=true2025.0383.8
KATHLEEN-CLEANParams=733K, Attention=false, Tokenization=false2026.0483.3
Task VectorBackbone=LLaMA-2-13B2026.0582.8
State VectorBackbone=LLaMA-2-7B2026.0582.56
Task VectorBackbone=LLaMA-2-7B2026.0581.6
USMoEBackbone=Transformer-XL(20M), Top-k=2, FLOPs=7.7620 x 10^10, Fine-tuning=true2025.0381.5
ECBackbone=Transformer-XL(20M), Routing=Expert Choice (EC), Top-k=2, FLOPs=7.7620 x 10^10, Fine-tuning=true2025.0381.5
Quaternion Attention (Tay et al., 2019)Latency (ms / batch)=3.92 ± 0.122026.0581.12
shared-score quaternion self-attention mechanismLatency (ms / batch)=2.87 ± 0.072026.0581.04
Zero-shotBackbone=LLaMA-2-7B2026.0580.4
I2CLBackbone=LLaMA-2-7B2026.0580.16
Kathleen FFT+PLGParams=149K, Attention=false, Tokenization=false2026.0478.6
Function VectorBackbone=LLaMA-2-7B2026.0578.6
SMoE-DRBackbone=Transformer-XL(20M), Routing=Token Choice (TC), Top-k=2, FLOPs=7.7620 x 10^10, Fine-tuning=true2025.0378.6
StableMoEBackbone=Transformer-XL(20M), Routing=Token Choice (TC), Top-k=2, FLOPs=7.7620 x 10^10, Fine-tuning=true2025.0377.7
SMoEBackbone=Transformer-XL(20M), Routing=Token Choice (TC), Top-k=2, FLOPs=7.7620 x 10^10, Fine-tuning=true2025.0377.1
XMoEBackbone=Transformer-XL(20M), Routing=Token Choice (TC), Top-k=2, FLOPs=7.7620 x 10^10, Fine-tuning=true2025.0376.7
Function VectorBackbone=LLaMA-2-13B2026.0572.44
Zero-shotBackbone=LLaMA-2-13B2026.0566.6
I2CLBackbone=LLaMA-2-13B2026.0564.8
Function VectorBackbone=Qwen-2.5-7B2026.0558.48
Zero-shotBackbone=Qwen-2.5-7B2026.0554.8