Community Alignment on Community Alignment (CA)
57.2AccuracyCUMA
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| CUMABackbone=Qwen3-8B, Category=Sparsely Activated Adapters, Trainable Params=4.15%, Rank (r)=642026.01 | 57.2 | 44.8 | 77.5 | 78.2 | |
| CUMABackbone=Qwen3-8B, Category=Sparsely Activated Adapters, Trainable Params=1.53%, Rank (r)=82026.01 | 55.4 | 43.1 | 75.8 | 76.5 | |
| HydraLoRABackbone=Qwen3-8B, Category=Sparsely Activated Adapters, Trainable Params=2.31%2026.01 | 52.8 | 40.2 | 71.5 | 73.6 | |
| CUMABackbone=Llama-3.1-8B, Category=Sparsely Activated Adapters, Trainable Params=4.15%, Rank (r)=642026.01 | 52.45 | 40.12 | 72.2 | 74.5 | |
| MixLoRABackbone=Qwen3-8B, Category=Sparsely Activated Adapters, Trainable Params=3.01%2026.01 | 51.5 | 38.8 | 70.5 | 72.8 | |
| CUMABackbone=Llama-3.1-8B, Category=Sparsely Activated Adapters, Trainable Params=1.53%, Rank (r)=82026.01 | 50.12 | 38.5 | 68.5 | 73.8 | |
| Full Fine-Tuning (FFT)Backbone=Qwen3-8B, Category=Dense Fine-Tuning, Trainable Params=100.0%2026.01 | 49.5 | 36.2 | 66.8 | 68.5 | |
| HydraLoRABackbone=Llama-3.1-8B, Category=Sparsely Activated Adapters, Trainable Params=2.31%2026.01 | 47.9 | 36.2 | 69.8 | 69.5 | |
| P-Tuning v2Backbone=Qwen3-8B, Category=Dense Fine-Tuning, Trainable Params=0.94%2026.01 | 47.5 | 34.8 | 59.5 | 61.2 | |
| MixLoRABackbone=Llama-3.1-8B, Category=Sparsely Activated Adapters, Trainable Params=3.01%2026.01 | 46.8 | 34.6 | 66.5 | 68.2 | |
| Full Fine-Tuning (FFT)Backbone=Llama-3.1-8B, Category=Dense Fine-Tuning, Trainable Params=100.0%2026.01 | 45.15 | 32.3 | 63.5 | 65.2 | |
| P-Tuning v2Backbone=Llama-3.1-8B, Category=Dense Fine-Tuning, Trainable Params=0.94%2026.01 | 43.5 | 30.85 | 57.2 | 58.8 | |
| DoRABackbone=Llama-3.1-8B, Category=Dense Fine-Tuning, Trainable Params=0.38%, Rank (r)=642026.01 | 39.2 | 31.5 | 61.8 | 63.5 | |
| DoRABackbone=Qwen3-8B, Category=Dense Fine-Tuning, Trainable Params=0.38%, Rank (r)=642026.01 | 39.2 | 31.5 | 64.5 | 66.8 | |
| LoRABackbone=Llama-3.1-8B, Category=Dense Fine-Tuning, Trainable Params=0.37%, Rank (r)=642026.01 | 38.53 | 30.5 | 60.5 | 62.1 | |
| LoRABackbone=Qwen3-8B, Category=Dense Fine-Tuning, Trainable Params=0.37%, Rank (r)=642026.01 | 38.53 | 30.5 | 63.2 | 65.5 | |
| Persona PromptingBackbone=Qwen3-8B, Category=Inference-Time Strategies, Trainable Params=0.00%2026.01 | 32.8 | 21 | 57.1 | 58.5 | |
| Vanilla BaselineBackbone=Qwen3-8B, Category=Inference-Time Strategies, Trainable Params=0.00%2026.01 | 31.2 | 17.75 | — | — | |
| Prompt Steering (3-shot)Backbone=Llama-3.1-8B, Category=Inference-Time Strategies, Trainable Params=0.00%2026.01 | 26.8 | 22.74 | 56.8 | 57.5 | |
| Vanilla BaselineBackbone=Llama-3.1-8B, Category=Inference-Time Strategies, Trainable Params=0.00%2026.01 | 26.7 | 20.79 | — | — | |
| Persona PromptingBackbone=Llama-3.1-8B, Category=Inference-Time Strategies, Trainable Params=0.00%2026.01 | 26.1 | 21.57 | 55.5 | 56.2 | |
| Prompt Steering (3-shot)Backbone=Qwen3-8B, Category=Inference-Time Strategies, Trainable Params=0.00%2026.01 | 26 | 22.19 | 59.5 | 60.8 |