Best feasible prompt identification on XSum
0.161Average Soft Constrained Reward (b=3)CSR
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| CSRBackbone Model=Llama3-8B, K (Prompt pool size)=302026.05 | 0.161 | 0.154 | 0.148 | 0.149 | |
| CSRBackbone Model=Llama3-8B, K (Prompt pool size)=502026.05 | 0.144 | 0.134 | 0.122 | 0.157 | |
| MLP-CSRBackbone Model=Gemma-7B, K (Prompt pool size)=502026.05 | 0.143 | 0.142 | 0.147 | 0.142 | |
| MLP-CSRBackbone Model=Llama3-8B, K (Prompt pool size)=302026.05 | 0.143 | 0.126 | 0.147 | 0.149 | |
| MLP-CSRBackbone Model=Llama3-8B, K (Prompt pool size)=502026.05 | 0.141 | 0.151 | 0.154 | 0.156 | |
| CSRBackbone Model=Llama3-8B, K (Prompt pool size)=1002026.05 | 0.141 | 0.144 | 0.153 | 0.154 | |
| MLP-CSRBackbone Model=Gemma-7B, K (Prompt pool size)=1002026.05 | 0.139 | 0.142 | 0.145 | 0.147 | |
| CSRBackbone Model=Gemma-7B, K (Prompt pool size)=1002026.05 | 0.134 | 0.141 | 0.142 | 0.143 | |
| MLP-CSRBackbone Model=Llama3-8B, K (Prompt pool size)=1002026.05 | 0.124 | 0.144 | 0.158 | 0.157 | |
| MLP-CSRBackbone Model=Gemma-7B, K (Prompt pool size)=302026.05 | 0.123 | 0.139 | 0.14 | 0.142 | |
| CSRBackbone Model=Gemma-7B, K (Prompt pool size)=502026.05 | 0.122 | 0.143 | 0.141 | 0.14 | |
| CSRBackbone Model=Gemma-7B, K (Prompt pool size)=302026.05 | 0.117 | 0.137 | 0.144 | 0.143 | |
| UniformBackbone Model=Llama3-8B, K (Prompt pool size)=1002026.05 | 0.113 | 0.134 | 0.123 | 0.141 | |
| UniformBackbone Model=Llama3-8B, K (Prompt pool size)=502026.05 | 0.063 | 0.045 | 0.051 | 0.078 | |
| UniformBackbone Model=Llama3-8B, K (Prompt pool size)=302026.05 | 0.048 | 0.069 | 0.08 | 0.089 | |
| UniformBackbone Model=Gemma-7B, K (Prompt pool size)=502026.05 | 0.021 | 0.021 | 0.037 | 0.036 | |
| UniformBackbone Model=Gemma-7B, K (Prompt pool size)=1002026.05 | 0.021 | 0.066 | 0.037 | 0.044 | |
| UniformBackbone Model=Gemma-7B, K (Prompt pool size)=302026.05 | 0.015 | 0 | 0.03 | 0.037 |