Text Classification on AGNews (Accuracy)
94.4AccuracyBERT
Evaluation results
| Method | Links | |
|---|---|---|
| BERTPre-training=true, Training=true2026.06 | 94.4 | |
| BERT-baseParams=110M, Attention=true, Tokenization=true2026.04 | 94 | |
| DistilBERTParams=66M, Attention=true, Tokenization=true2026.04 | 93 | |
| SPAModel=Qwen2.5-14B-it2026.04 | 92.5 | |
| KATHLEEN-CLEANParams=733K, Attention=false, Tokenization=false2026.04 | 92.3 | |
| RSN-TuneModel=Qwen2.5-14B-it2026.04 | 92.2 | |
| RandomModel=Qwen2.5-14B-it2026.04 | 91.9 | |
| SPAModel=Qwen2.5-7B-it2026.04 | 90.8 | |
| UpriseModel=LLaMA3-8B, Setting=Few-shot2026.05 | 90.7 | |
| RSN-TuneModel=Qwen2.5-7B-it2026.04 | 90.6 | |
| SPAModel=Llama3-8B-it2026.04 | 90.5 | |
| Tok. KathleenParams=11.8M, Attention=false, Tokenization=true2026.04 | 90.2 | |
| RSN-TuneModel=Llama3-8B-it2026.04 | 90.2 | |
| RandomModel=Qwen2.5-7B-it2026.04 | 90.1 | |
| RandomModel=Llama3-8B-it2026.04 | 89.8 | |
| DiSPModel=LLaMA3-8B, Setting=Few-shot2026.05 | 89.7 | |
| UpriseModel=Qwen2.5-7B, Setting=Few-shot2026.05 | 89.4 | |
| DiSPModel=Qwen2.5-7B, Setting=Few-shot2026.05 | 89.4 | |
| FedAFDData Setting=IID, Evaluation Location=Clients, #rounds=212026.03 | 89.34 | |
| Kathleen FFT+PLGParams=149K, Attention=false, Tokenization=false2026.04 | 89.3 | |
| W2VPre-training=true, Training=false2026.06 | 89.2 | |
| DCModel=Qwen, Shots=42025.05 | 88.91 | |
| LDicePre-training=false, Training=false, k=72026.06 | 88.5 | |
| LJaccardPre-training=false, Training=false, k=72026.06 | 88.5 | |
| SeDPOModel=LLaMA3-8B, Setting=Few-shot2026.05 | 88.1 | |
| Se2Model=Qwen2.5-7B, Setting=Few-shot2026.05 | 88.1 | |
| ICLBackbone=LLaMA-2-13B2026.05 | 88.04 | |
| NCDgzipPre-training=false, Training=false, k=72026.06 | 87.6 | |
| NCDlpPre-training=false, Training=false, k=72026.06 | 87.6 | |
| SeDPOModel=Qwen2.5-7B, Setting=Few-shot2026.05 | 87.4 | |
| ICLBackbone=LLaMA-3.1-8B2026.05 | 87.16 | |
| LTVBackbone=LLaMA-2-13B2026.05 | 86.68 | |
| ICLBackbone=Qwen-3-8B2026.05 | 86.56 | |
| BM25Model=LLaMA3-8B, Setting=Few-shot2026.05 | 86.1 | |
| Se2Model=LLaMA3-8B, Setting=Few-shot2026.05 | 86.1 | |
| LSTMPre-training=false, Training=true2026.06 | 86.1 | |
| FedETData Setting=IID, Evaluation Location=Clients, #rounds=232026.03 | 86.07 | |
| FedGEMSData Setting=IID, Evaluation Location=Clients, #rounds=262026.03 | 85.63 | |
| LOCALData Setting=IID, Evaluation Location=Clients, #rounds=292026.03 | 85.56 | |
| FedMDData Setting=IID, Evaluation Location=Clients, #rounds=252026.03 | 85.54 | |
| CCModel=Qwen, Shots=42025.05 | 85.47 | |
| ICLBackbone=Qwen-2.5-7B2026.05 | 85.4 | |
| SafeInstrSample number (n)=1000, Harmful ratio (p)=0.15, Setting=default2025.06 | 85.4 | |
| CreamFLData Setting=IID, Evaluation Location=Clients, #rounds=222026.03 | 85.22 | |
| FedDFAData Setting=IID, Evaluation Location=Clients, #rounds=252026.03 | 84.83 | |
| State VectorBackbone=LLaMA-2-13B2026.05 | 84.64 | |
| FedMKDData Setting=IID, Evaluation Location=Clients, #rounds=222026.03 | 84.63 | |
| AsFTSample number (n)=1000, Harmful ratio (p)=0.15, Setting=default2025.06 | 84.5 | |
| ICLBackbone=LLaMA-2-7B2026.05 | 84.48 | |
| BEASample number (n)=1000, Harmful ratio (p)=0.1, Setting=default2025.06 | 84.4 | |
| SFTSample number (n)=1000, Harmful ratio (p)=0.1, Setting=default2025.06 | 84.3 | |
| SFTSample number (n)=1000, Harmful ratio (p)=0.15, Setting=default2025.06 | 84.3 | |
| SafeInstrSample number (n)=1000, Harmful ratio (p)=0.1, Setting=default2025.06 | 84.3 | |
| AsFTSample number (n)=1000, Harmful ratio (p)=0.05, Setting=default2025.06 | 84.3 | |
| AsFTSample number (n)=1000, Harmful ratio (p)=0.1, Setting=default2025.06 | 84.3 | |
| RandomModel=LLaMA3-8B, Setting=Few-shot2026.05 | 84.1 | |
| BM25Model=Qwen2.5-7B, Setting=Few-shot2026.05 | 84.1 | |
| LTVBackbone=LLaMA-2-7B2026.05 | 83.96 | |
| BaseModel=Qwen2.5-14B-it2026.04 | 83.9 | |
| SafeInstrSample number (n)=1000, Harmful ratio (p)=clean, Setting=default2025.06 | 83.9 | |
| SafeInstrSample number (n)=1000, Harmful ratio (p)=Average, Setting=default2025.06 | 83.86 | |
| SFTSample number (n)=1000, Harmful ratio (p)=0.2, Setting=default2025.06 | 83.8 | |
| SafeInstrSample number (n)=1000, Harmful ratio (p)=0.2, Setting=default2025.06 | 83.8 | |
| AsFTSample number (n)=1000, Harmful ratio (p)=Average, Setting=default2025.06 | 83.78 | |
| SFTSample number (n)=1000, Harmful ratio (p)=Average, Setting=default2025.06 | 83.26 | |
| AsFTSample number (n)=1000, Harmful ratio (p)=clean, Setting=default2025.06 | 83 | |
| SFTSample number (n)=1000, Harmful ratio (p)=clean, Setting=default2025.06 | 82.9 | |
| Safe LoRASample number (n)=1000, Harmful ratio (p)=clean, Setting=default2025.06 | 82.9 | |
| LTVBackbone=LLaMA-3.1-8B2026.05 | 82.84 | |
| AsFTSample number (n)=1000, Harmful ratio (p)=0.2, Setting=default2025.06 | 82.8 | |
| BEASample number (n)=1000, Harmful ratio (p)=clean, Setting=default2025.06 | 82.6 | |
| Base LLMModel=Llama, Shots=42025.05 | 82.58 | |
| Base LLMModel=Mistral, Shots=42025.05 | 82.5 | |
| Lisa-alignedSample number (n)=1000, Harmful ratio (p)=clean, Setting=default2025.06 | 82.4 | |
| Safe LoRASample number (n)=1000, Harmful ratio (p)=0.15, Setting=default2025.06 | 82.2 | |
| Lisa-alignedSample number (n)=1000, Harmful ratio (p)=0.15, Setting=default2025.06 | 82 | |
| SafeInstrSample number (n)=1000, Harmful ratio (p)=0.05, Setting=default2025.06 | 81.9 | |
| Lisa-alignedSample number (n)=1000, Harmful ratio (p)=0.1, Setting=default2025.06 | 81.8 | |
| TextCNNPre-training=false, Training=true2026.06 | 81.7 | |
| Task VectorBackbone=LLaMA-3.1-8B2026.05 | 81.36 | |
| State VectorBackbone=Qwen-3-8B2026.05 | 81.24 | |
| Safe LoRASample number (n)=1000, Harmful ratio (p)=0.1, Setting=default2025.06 | 81.2 | |
| SFTSample number (n)=1000, Harmful ratio (p)=0.05, Setting=default2025.06 | 81 | |
| BEASample number (n)=1000, Harmful ratio (p)=0.15, Setting=default2025.06 | 81 | |
| Safe LoRASample number (n)=1000, Harmful ratio (p)=Average, Setting=default2025.06 | 80.98 | |
| RandomModel=Qwen2.5-7B, Setting=Few-shot2026.05 | 80.4 | |
| State VectorBackbone=LLaMA-3.1-8B2026.05 | 80.28 | |
| SCModel=Mistral, Shots=42025.05 | 80.23 | |
| Safe LoRASample number (n)=1000, Harmful ratio (p)=0.2, Setting=default2025.06 | 80 | |
| Lisa-alignedSample number (n)=1000, Harmful ratio (p)=Average, Setting=default2025.06 | 79.94 | |
| Task VectorBackbone=LLaMA-2-13B2026.05 | 79.84 | |
| DCModel=Llama, Shots=42025.05 | 79.61 | |
| BaseModel=Qwen2.5-7B-it2026.04 | 79.6 | |
| LossBackbone=Llama-3.1-8B2026.03 | 79.4 | |
| CCModel=Llama, Shots=42025.05 | 79.3 | |
| Task VectorBackbone=Qwen-3-8B2026.05 | 79.24 | |
| BEASample number (n)=1000, Harmful ratio (p)=Average, Setting=default2025.06 | 79.08 | |
| I2CLBackbone=LLaMA-2-13B2026.05 | 78.88 | |
| MSSRfullBackbone=Llama-3.1-8B2026.03 | 78.7 | |
| Safe LoRASample number (n)=1000, Harmful ratio (p)=0.05, Setting=default2025.06 | 78.6 |