Multi-Task Instruct-Tuning on SuperNI (test)
57.35ROUGE ScoreBADIT
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| BADITBackbone=Gemma2-9B, Implementation Setting=Mixed Training2026.05 | 57.35 | 0.54 | — | — | |
| BADITBackbone=Qwen3-8B, Implementation Setting=Mixed Training2026.05 | 55.87 | 0.92 | — | — | |
| BADITBackbone=Qwen3-4B, Implementation Setting=Mixed Training2026.05 | 55.22 | 0.91 | — | — | |
| LoRABackbone=Gemma2-9B, Implementation Setting=Mixed Training2026.05 | 55.16 | -1.35 | — | — | |
| OLoRABackbone=Gemma2-9B, Implementation Setting=Mixed Training2026.05 | 54.78 | -1.05 | — | — | |
| LoRAMoEBackbone=Gemma2-9B, Implementation Setting=Mixed Training2026.05 | 54.78 | -0.91 | — | — | |
| BADITBackbone=Llama3-8B, Implementation Setting=Mixed Training2026.05 | 54.75 | 0.79 | — | — | |
| LoRAMoEBackbone=Qwen3-8B, Implementation Setting=Mixed Training2026.05 | 54.64 | -0.57 | — | — | |
| OLoRABackbone=Qwen3-8B, Implementation Setting=Mixed Training2026.05 | 54.53 | -0.45 | — | — | |
| GainLoRABackbone=Qwen3-8B, Implementation Setting=Mixed Training2026.05 | 54.33 | -1.78 | — | — | |
| GainLoRABackbone=Gemma2-9B, Implementation Setting=Mixed Training2026.05 | 54.29 | -1.88 | — | — | |
| LoRABackbone=Qwen3-8B, Implementation Setting=Mixed Training2026.05 | 54.22 | -1.66 | — | — | |
| PiSSABackbone=Gemma2-9B, Implementation Setting=Mixed Training2026.05 | 54.2 | -1.78 | — | — | |
| LoRAMoEBackbone=Qwen3-4B, Implementation Setting=Mixed Training2026.05 | 53.63 | -0.68 | — | — | |
| BADITBackbone=Gemma2-2B, Implementation Setting=Mixed Training2026.05 | 53.6 | 0.7 | — | — | |
| PiSSABackbone=Qwen3-8B, Implementation Setting=Mixed Training2026.05 | 53.57 | -2.3 | — | — | |
| OLoRABackbone=Qwen3-4B, Implementation Setting=Mixed Training2026.05 | 53.5 | 0.15 | — | — | |
| LoRABackbone=Qwen3-4B, Implementation Setting=Mixed Training2026.05 | 53.31 | -1.24 | — | — | |
| GainLoRABackbone=Qwen3-4B, Implementation Setting=Mixed Training2026.05 | 53.24 | -1.33 | — | — | |
| LoRAMoEBackbone=Llama3-8B, Implementation Setting=Mixed Training2026.05 | 52.95 | -1.96 | — | — | |
| OLoRABackbone=Llama3-8B, Implementation Setting=Mixed Training2026.05 | 52.89 | -0.41 | — | — | |
| PiSSABackbone=Llama3-8B, Implementation Setting=Mixed Training2026.05 | 52.88 | -0.72 | — | — | |
| PiSSABackbone=Qwen3-4B, Implementation Setting=Mixed Training2026.05 | 52.87 | -0.45 | — | — | |
| GainLoRABackbone=Llama3-8B, Implementation Setting=Mixed Training2026.05 | 52.71 | -2.13 | — | — | |
| LoRABackbone=Llama3-8B, Implementation Setting=Mixed Training2026.05 | 52.58 | -1.02 | — | — | |
| OLoRABackbone=Gemma2-2B, Implementation Setting=Mixed Training2026.05 | 51.82 | -0.87 | — | — | |
| LoRABackbone=Gemma2-2B, Implementation Setting=Mixed Training2026.05 | 51.72 | -1.3 | — | — | |
| BADITBackbone=Llama3-3B, Implementation Setting=Mixed Training2026.05 | 51.45 | 0.61 | — | — | |
| LoRAMoEBackbone=Gemma2-2B, Implementation Setting=Mixed Training2026.05 | 51.36 | -1.05 | — | — | |
| GainLoRABackbone=Gemma2-2B, Implementation Setting=Mixed Training2026.05 | 51.26 | -1.7 | — | — | |
| PiSSABackbone=Gemma2-2B, Implementation Setting=Mixed Training2026.05 | 50.96 | -1.28 | — | — | |
| BADITBackbone=Qwen3-8B, Implementation Setting=Sequential Training2026.05 | 50.86 | 0.44 | 6.95 | -5.43 | |
| BADITBackbone=Qwen3-4B, Implementation Setting=Sequential Training2026.05 | 50.74 | 0.62 | 5.95 | -4.21 | |
| LoRABackbone=Llama3-3B, Implementation Setting=Mixed Training2026.05 | 50.07 | -0.23 | — | — | |
| OLoRABackbone=Llama3-3B, Implementation Setting=Mixed Training2026.05 | 49.83 | -0.47 | — | — | |
| LoRAMoEBackbone=Llama3-3B, Implementation Setting=Mixed Training2026.05 | 49.72 | -0.59 | — | — | |
| PiSSABackbone=Llama3-3B, Implementation Setting=Mixed Training2026.05 | 49.64 | -0.27 | — | — | |
| GainLoRABackbone=Llama3-3B, Implementation Setting=Mixed Training2026.05 | 49.6 | -0.23 | — | — | |
| BADITBackbone=Llama3-8B, Implementation Setting=Sequential Training2026.05 | 48.83 | 1.71 | 8.57 | -3.49 | |
| LoRAMoEBackbone=Qwen3-4B, Implementation Setting=Sequential Training2026.05 | 48.58 | -0.74 | 8.5 | -6.25 | |
| OLoRABackbone=Qwen3-8B, Implementation Setting=Sequential Training2026.05 | 48.57 | -1.15 | 8.73 | -6.78 | |
| LoRABackbone=Qwen3-4B, Implementation Setting=Sequential Training2026.05 | 48.54 | -1.31 | 8.53 | -6.68 | |
| GainLoRABackbone=Qwen3-8B, Implementation Setting=Sequential Training2026.05 | 48.44 | -0.73 | 8.96 | -6.42 | |
| BADITBackbone=Gemma2-9B, Implementation Setting=Sequential Training2026.05 | 48.34 | 0.39 | 10.31 | -8.45 | |
| PiSSABackbone=Qwen3-4B, Implementation Setting=Sequential Training2026.05 | 48.26 | -0.64 | 8.05 | -6.12 | |
| BADITBackbone=Llama3-3B, Implementation Setting=Sequential Training2026.05 | 48.13 | 0.71 | 6.25 | -2.57 | |
| LoRAMoEBackbone=Qwen3-8B, Implementation Setting=Sequential Training2026.05 | 48.07 | -1.26 | 8.47 | -6.23 | |
| GainLoRABackbone=Qwen3-4B, Implementation Setting=Sequential Training2026.05 | 48.07 | -1.26 | 8.47 | -6.23 | |
| OLoRABackbone=Qwen3-4B, Implementation Setting=Sequential Training2026.05 | 47.83 | -1.77 | 8.78 | -6.47 | |
| PiSSABackbone=Qwen3-8B, Implementation Setting=Sequential Training2026.05 | 47.69 | -0.06 | 7.5 | -6.83 | |
| LoRABackbone=Qwen3-8B, Implementation Setting=Sequential Training2026.05 | 47.08 | -0.11 | 9.21 | -8.11 | |
| LoRAMoEBackbone=Gemma2-9B, Implementation Setting=Sequential Training2026.05 | 46.98 | 0.17 | 11.99 | -10.11 | |
| LoRABackbone=Gemma2-9B, Implementation Setting=Sequential Training2026.05 | 46.8 | -0.32 | 10.46 | -9.38 | |
| BADITBackbone=Gemma2-2B, Implementation Setting=Sequential Training2026.05 | 46.76 | 0.25 | 9.3 | -7.65 | |
| GainLoRABackbone=Gemma2-9B, Implementation Setting=Sequential Training2026.05 | 46.51 | -0.62 | 12.62 | -10.8 | |
| LoRAMoEBackbone=Llama3-3B, Implementation Setting=Sequential Training2026.05 | 46.2 | -0.59 | 8.58 | -3.51 | |
| OLoRABackbone=Gemma2-9B, Implementation Setting=Sequential Training2026.05 | 46.05 | -0.69 | 11.2 | -13.57 | |
| OLoRABackbone=Llama3-3B, Implementation Setting=Sequential Training2026.05 | 45.59 | -0.74 | 10.07 | -5.95 | |
| PiSSABackbone=Llama3-3B, Implementation Setting=Sequential Training2026.05 | 45.49 | -0.93 | 9.13 | -5.27 | |
| LoRAMoEBackbone=Llama3-8B, Implementation Setting=Sequential Training2026.05 | 45.47 | 0.27 | 11.78 | -9.71 | |
| GainLoRABackbone=Llama3-3B, Implementation Setting=Sequential Training2026.05 | 45.34 | -0.78 | 9.67 | -5.66 | |
| OLoRABackbone=Llama3-8B, Implementation Setting=Sequential Training2026.05 | 45.25 | 0 | 11.45 | -10.01 | |
| PiSSABackbone=Llama3-8B, Implementation Setting=Sequential Training2026.05 | 45.23 | -0.63 | 12.54 | -10.67 | |
| LoRABackbone=Llama3-3B, Implementation Setting=Sequential Training2026.05 | 45.14 | -0.17 | 9.79 | -4.91 | |
| GainLoRABackbone=Llama3-8B, Implementation Setting=Sequential Training2026.05 | 45.04 | -1.17 | 12.7 | -10.94 | |
| PiSSABackbone=Gemma2-9B, Implementation Setting=Sequential Training2026.05 | 44.93 | -2.24 | 12.27 | -14.8 | |
| OLoRABackbone=Gemma2-2B, Implementation Setting=Sequential Training2026.05 | 44.93 | -1.03 | 10.94 | -10.55 | |
| LoRABackbone=Llama3-8B, Implementation Setting=Sequential Training2026.05 | 44.88 | -1.9 | 12.41 | -10.23 | |
| GainLoRABackbone=Gemma2-2B, Implementation Setting=Sequential Training2026.05 | 44.16 | -0.92 | 11.81 | -10.27 | |
| LoRAMoEBackbone=Gemma2-2B, Implementation Setting=Sequential Training2026.05 | 43.88 | -0.86 | 11.66 | -9.95 | |
| LoRABackbone=Gemma2-2B, Implementation Setting=Sequential Training2026.05 | 43.16 | -0.03 | 13.21 | -11.28 | |
| PiSSABackbone=Gemma2-2B, Implementation Setting=Sequential Training2026.05 | 42.57 | -1.65 | 12.74 | -10.57 |