Multi-objective Reinforcement Learning on RLVR-GSM
0.0112Multiplicative Gap (ε)PALM
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| PALMSize=42026.04 | 0.0112 | 0.0046 | |
| PALMSize=32026.04 | 0.0193 | 0.0081 | |
| RandomSize=42026.04 | 0.0872 | 0.0482 | |
| UniformSize=42026.04 | 0.0915 | 0.0499 | |
| PALMSize=22026.04 | 0.104 | 0.0319 | |
| RandomSize=32026.04 | 0.124 | 0.074 | |
| UniformSize=32026.04 | 0.1363 | 0.0784 | |
| RandomSize=22026.04 | 0.1742 | 0.106 | |
| UniformSize=22026.04 | 0.188 | 0.1143 | |
| PALMSize=12026.04 | 0.2767 | 0.2026 | |
| RandomSize=12026.04 | 0.2842 | 0.1698 | |
| UniformSize=12026.04 | 0.2962 | 0.1841 |