Online Self-Weighted Fine-Tuning · CiteArk

Recent activity

—Preparing data, models and dependencies
—Orchestrating training, evaluation and analysis
—Orchestrating training, evaluation and analysis
—Analysis environment started
—Execution needs attention

Reproduction materials

Prepared 0/0 materials

Links and access are readyScientific material completeness undetermined

资料与依赖由同一实验工作区准备，直接复用已有缓存，不另启准备 Agent。

This experiment has no separate content download, but dependencies, configuration, and loaders are still checked.

Prefetching only saves time. Unprepared material is handed to the complete reproduction Agent to download, adapt, and verify autonomously.

Credit estimate for this reproduction

Expected usage

509

Maximum reservation

1323

Estimated compute time

About 195 min

Actual usage was 156 credits; 0 credits were charged. Compute ran for about 9 minutes and the model used 6209368 tokens.

Online Self-Weighted Fine-Tuning

Public

Authors:Haiquan Wen, Yiwei He, Bei Peng, Guangliang Cheng

arXiv 2026 Machine Learning (cs.LG)Artificial Intelligence

Reproduction results

All claims ↗

No reproduction results yet

The paper’s claims are available in Research claims.

Full research claim catalog

0 / 6 claims verified

the rest still being verified

Key metricsPaper-reported → reproduced · click a row to expand

In rollout count ablation on Qwen3-0.6B-Base, increasing K from 1 to 2 yields a significant performance jump (from 15.15% to 20.20% on GPQA-Diamond, and from 6.67% to 10.00% on AIME 25 Pass@16), while further increasing K to 4 or 8 yields negligible marginal benefits.Whole claim: Experiment blocked; see the specific reason0/10 measurements supported · 0 assessed10measurementsExplore results

The rollout ablation K=1 vs K=2 tests the core theoretical inflection point on Qwen3-0.6B-Base and can be partially evaluated alongside the main experiment.

Measurement results

0/10 assessed

0 supported0 disputed or conflicting10 not assessed

The complete result group stays with its claim. Assessed rows come first; their judgments do not replace the whole-claim conclusion. Expand a row for conditions, sources, and evidence history.

Each row is a measurement with its reported value, latest available observation, and cumulative evidence state.
Reported	Latest observation	Evidence judgment
6.67 percentage_points	—	Not assessed
6.67 percentage_points	—	Not assessed
22.8 percentage_points	—	Not assessed
15.15 percentage_points	—	Not assessed
6.67 percentage_points	—	Not assessed
10 percentage_points	—	Not assessed
24.2 percentage_points	—	Not assessed
20.2 percentage_points	—	Not assessed
20.71 percentage_points	—	Not assessed
20.71 percentage_points	—	Not assessed

Open claim details

OSW-FT (K=2) on Qwen3-1.7B-Base achieves 5.42% Pass@1 and 26.67% Pass@16 on AIME 24, 3.96% Pass@1 and 23.34% Pass@16 on AIME 25, 22.24% Pass@1 and 55.81% Pass@16 on AMC 22, 27.45% Pass@1 and 65.22% Pass@16 on AMC 23, 17.78% Pass@1 and 48.89% Pass@16 on AMC 24, 57.40% Pass@1 on MATH-500, and 26.27% Pass@1 on GPQA-Diamond.Whole claim: Experiment blocked; see the specific reason0/6 measurements supported · 0 assessed6measurementsExplore results

Training Qwen3-1.7B full-parameter fine-tuning for 1 epoch with 2 online rollouts exceeds the cumulative compute and runtime budget when executed alongside the primary 0.6B verification experiments.

Measurement results

0/6 assessed

0 supported0 disputed or conflicting6 not assessed

The complete result group stays with its claim. Assessed rows come first; their judgments do not replace the whole-claim conclusion. Expand a row for conditions, sources, and evidence history.

Each row is a measurement with its reported value, latest available observation, and cumulative evidence state.
Reported	Latest observation	Evidence judgment
5.42 percentage_points	—	Not assessed
26.67 percentage_points	—	Not assessed
3.96 percentage_points	—	Not assessed
23.34 percentage_points	—	Not assessed
57.4 percentage_points	—	Not assessed
26.27 percentage_points	—	Not assessed

Open claim details

OSW-FT (K=2) on Qwen3-0.6B-Base achieves 0.83% Pass@1 and 6.67% Pass@16 on AIME 24, 1.04% Pass@1 and 10.00% Pass@16 on AIME 25, 8.72% Pass@1 and 41.86% Pass@16 on AMC 22, 13.04% Pass@1 and 56.52% Pass@16 on AMC 23, 5.14% Pass@1 and 28.89% Pass@16 on AMC 24, 24.20% Pass@1 on MATH-500, and 20.20% Pass@1 on GPQA-Diamond, matching or outperforming standard SFT across all evaluated benchmarks.Whole claim: Awaiting reproduction0/12 measurements supported · 0 assessed12measurementsExplore results

Experiment plan ready; no runs yet.

Measurement results

0/12 assessed

0 supported0 disputed or conflicting12 not assessed

The complete result group stays with its claim. Assessed rows come first; their judgments do not replace the whole-claim conclusion. Expand a row for conditions, sources, and evidence history.

Each row is a measurement with its reported value, latest available observation, and cumulative evidence state.
Reported	Latest observation	Evidence judgment
0.83 percentage_points	—	Not assessed
6.67 percentage_points	—	Not assessed
1.04 percentage_points	—	Not assessed
10 percentage_points	—	Not assessed
8.72 percentage_points	—	Not assessed
41.86 percentage_points	—	Not assessed
13.04 percentage_points	—	Not assessed
56.52 percentage_points	—	Not assessed
5.14 percentage_points	—	Not assessed
28.89 percentage_points	—	Not assessed
24.2 percentage_points	—	Not assessed
20.2 percentage_points	—	Not assessed

Open claim details

OSW-FT (K=2) on Qwen3-4B-Base achieves 13.33% Pass@1 and 36.67% Pass@16 on AIME 24, 16.25% Pass@1 and 36.67% Pass@16 on AIME 25, 41.42% Pass@1 and 74.42% Pass@16 on AMC 22, 45.11% Pass@1 and 78.26% Pass@16 on AMC 23, 35.00% Pass@1 and 75.56% Pass@16 on AMC 24, 79.00% Pass@1 on MATH-500, and 32.32% Pass@1 on GPQA-Diamond.Whole claim: Experiment blocked; see the specific reason0/6 measurements supported · 0 assessed6measurementsExplore results

Training Qwen3-4B full-parameter fine-tuning for 1 epoch with 2 online rollouts requires multi-GPU distributed setup and exceeds the single L4 GPU memory and budget limits.

Measurement results

0/6 assessed

0 supported0 disputed or conflicting6 not assessed

The complete result group stays with its claim. Assessed rows come first; their judgments do not replace the whole-claim conclusion. Expand a row for conditions, sources, and evidence history.

Each row is a measurement with its reported value, latest available observation, and cumulative evidence state.
Reported	Latest observation	Evidence judgment
13.33 percentage_points	—	Not assessed
36.67 percentage_points	—	Not assessed
16.25 percentage_points	—	Not assessed
36.67 percentage_points	—	Not assessed
79 percentage_points	—	Not assessed
32.32 percentage_points	—	Not assessed

Open claim details

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).Whole claim: Experiment blocked; see the specific reason0/6 measurements supported · 0 assessed6measurementsExplore results

Running additional training runs at K=4 across multiple alternative weighting heuristics exceeds the cumulative paper budget allocation.

Measurement results

0/6 assessed

0 supported0 disputed or conflicting6 not assessed

The complete result group stays with its claim. Assessed rows come first; their judgments do not replace the whole-claim conclusion. Expand a row for conditions, sources, and evidence history.

Each row is a measurement with its reported value, latest available observation, and cumulative evidence state.
Reported	Latest observation	Evidence judgment
23.8 percentage_points	—	Not assessed
15.15 percentage_points	—	Not assessed
21.4 percentage_points	—	Not assessed
15.17 percentage_points	—	Not assessed
24 percentage_points	—	Not assessed
20.71 percentage_points	—	Not assessed

Open claim details

Other claims1

OSW-FT currently requires a reliable binary outcome verifier and ground-truth expert trajectories, limiting direct application to non-binary or open-ended generative tasks.Stated by the authors

This is a qualitative scope limitation acknowledged by the authors rather than an empirical numerical benchmark claim.

Open claim details