Online Self-Weighted Fine-Tuning
尚未评估计划受阻研究发现claim-weighting-strategy-ablation
Under K=4 rollouts on Qwen3-0.6B-Base, continuous online success-rate weighting in OSW-FT (24.00% MATH-500, 20.71% GPQA-Diamond) outperforms hard-thresholded binary rejection weighting (23.80% MATH-500, 15.15% GPQA-Diamond) and mean-matched random weighting (21.40% MATH-500, 15.17% GPQA-Diamond).
来源：source-paper:Table 3, Section 4.4, page 6
报告指标与观测值
pass_at_1
m-ablation-hard-math500-p1
论文报告 23.8 percentage_points
实际观测 — percentage_points
pass_at_1
m-ablation-hard-gpqad-p1
论文报告 15.15 percentage_points
实际观测 — percentage_points
pass_at_1
m-ablation-random-math500-p1
论文报告 21.4 percentage_points
实际观测 — percentage_points
pass_at_1
m-ablation-random-gpqad-p1
论文报告 15.17 percentage_points
实际观测 — percentage_points
pass_at_1
m-ablation-osw-math500-p1
论文报告 24 percentage_points
实际观测 — percentage_points
pass_at_1
m-ablation-osw-gpqad-p1
论文报告 20.71 percentage_points
实际观测 — percentage_points
Assessment（0）
这条 Claim 暂无已发布的不可变 Assessment。
关联运行（0）
这条 Claim 暂无关联执行。