β-OPSD Boosts Qwen3 Reasoning Performance, Ablation Reveals Key Mechanisms

Applying β-OPSD to Qwen3-1.7B increased its AIME 2024 score by 9.16 points. Ablation studies confirm that both a smoother target path and future-oriented credit assignment are crucial for the performance gains.

2026-08-04 ~ 2026-08-04 · 2 related posts

• β-OPSD lifts Qwen3-1.7B by 9.16 points on AIME 2024 avg@12 — furongh · 2026-08-04
• β-OPSD ablations show target path and return-to-go both drive gains — furongh · 2026-08-04