Soft constrained reward optimization on XSum on Gemma-7B
0.147Average Soft Constrained RewardMLP-CSR
Evaluation Results
| Method | Links | |
|---|---|---|
| MLP-CSRPrompt set size (K)=50, Per-arm budget (b)=82026.05 | 0.147 | |
| MLP-CSRPrompt set size (K)=100, Per-arm budget (b)=102026.05 | 0.147 | |
| MLP-CSRPrompt set size (K)=100, Per-arm budget (b)=82026.05 | 0.145 | |
| CSRPrompt set size (K)=30, Per-arm budget (b)=82026.05 | 0.144 | |
| CSRPrompt set size (K)=30, Per-arm budget (b)=102026.05 | 0.143 | |
| CSRPrompt set size (K)=50, Per-arm budget (b)=52026.05 | 0.143 | |
| MLP-CSRPrompt set size (K)=50, Per-arm budget (b)=32026.05 | 0.143 | |
| CSRPrompt set size (K)=100, Per-arm budget (b)=102026.05 | 0.143 | |
| MLP-CSRPrompt set size (K)=30, Per-arm budget (b)=102026.05 | 0.142 | |
| MLP-CSRPrompt set size (K)=50, Per-arm budget (b)=52026.05 | 0.142 | |
| MLP-CSRPrompt set size (K)=50, Per-arm budget (b)=102026.05 | 0.142 | |
| CSRPrompt set size (K)=100, Per-arm budget (b)=82026.05 | 0.142 | |
| MLP-CSRPrompt set size (K)=100, Per-arm budget (b)=52026.05 | 0.142 | |
| CSRPrompt set size (K)=50, Per-arm budget (b)=82026.05 | 0.141 | |
| CSRPrompt set size (K)=100, Per-arm budget (b)=52026.05 | 0.141 | |
| MLP-CSRPrompt set size (K)=30, Per-arm budget (b)=82026.05 | 0.14 | |
| CSRPrompt set size (K)=50, Per-arm budget (b)=102026.05 | 0.14 | |
| MLP-CSRPrompt set size (K)=30, Per-arm budget (b)=52026.05 | 0.139 | |
| MLP-CSRPrompt set size (K)=100, Per-arm budget (b)=32026.05 | 0.139 | |
| CSRPrompt set size (K)=30, Per-arm budget (b)=52026.05 | 0.137 | |
| CSRPrompt set size (K)=100, Per-arm budget (b)=32026.05 | 0.134 | |
| MLP-CSRPrompt set size (K)=30, Per-arm budget (b)=32026.05 | 0.123 | |
| CSRPrompt set size (K)=50, Per-arm budget (b)=32026.05 | 0.122 | |
| CSRPrompt set size (K)=30, Per-arm budget (b)=32026.05 | 0.117 | |
| UniformPrompt set size (K)=100, Per-arm budget (b)=52026.05 | 0.066 | |
| UniformPrompt set size (K)=100, Per-arm budget (b)=102026.05 | 0.044 | |
| UniformPrompt set size (K)=30, Per-arm budget (b)=102026.05 | 0.037 | |
| UniformPrompt set size (K)=50, Per-arm budget (b)=82026.05 | 0.037 | |
| UniformPrompt set size (K)=100, Per-arm budget (b)=82026.05 | 0.037 | |
| UniformPrompt set size (K)=50, Per-arm budget (b)=102026.05 | 0.036 | |
| UniformPrompt set size (K)=30, Per-arm budget (b)=82026.05 | 0.03 | |
| UniformPrompt set size (K)=50, Per-arm budget (b)=32026.05 | 0.021 | |
| UniformPrompt set size (K)=50, Per-arm budget (b)=52026.05 | 0.021 | |
| UniformPrompt set size (K)=100, Per-arm budget (b)=32026.05 | 0.021 | |
| UniformPrompt set size (K)=30, Per-arm budget (b)=32026.05 | 0.015 | |
| UniformPrompt set size (K)=30, Per-arm budget (b)=52026.05 | 0 |