Mathematical Reasoning on MATH L4
49.31Mean@16ME-ICPO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ME-ICPOBackbone=Qwen2.5-Math-7B, ME-ICPO=true2026.03 | 49.31 | 40.84 | — | |
| ME-ICPOBackbone=Qwen2.5-Math-1.5B, ME-ICPO=true2026.03 | 46.44 | 41.55 | — | |
| Qwen2.5-Math-7BME-ICPO=false2026.03 | 43.6 | 37.79 | — | |
| Qwen2.5-Math-1.5BME-ICPO=false2026.03 | 40.28 | 34.97 | — | |
| DreamBackbone=Dream-7B-Instruct, TPS=26.38, Speedup=1.0×2026.03 | — | — | 33.94 | |
| DualCacheBackbone=Dream-7B-Instruct, TPS=86.38, Speedup=3.3×2026.03 | — | — | 33.6 | |
| ES-dLLMBackbone=Dream-7B-Instruct, TPS=147.44, Speedup=5.6×2026.03 | — | — | 33.44 |