aiqbench analysis
FlowCTS introduces an on-policy distillation method specifically tailored for flow models, addressing exposure bias through continuous trajectory supervision rather than standard KL-divergence. By leveraging the integral relation between trajectories and velocity fields, this approach achieves faster convergence and measurable gains in generative evaluation metrics compared to vanilla baselines. The study highlights a critical trade-off where increasing supervision steps enriches information but simultaneously raises optimization difficulty for these complex models.
FlowCTS: On-policy Continuous Trajectory Supervision of Flow Models
arXiv:2607.24522v1 Announce Type: new Abstract: While on-policy distillation (OPD) effectively addresses sparse rewards and exposure bias in large language model post-training, its extension to flow models remains underexplored. To this end, we propose Flow Continuous Trajectory Supervision (FlowCTS), which matches subsequent student and reference trajectories initialized from the same student-visited state. Using the integral relation between trajectories and velocity fields, we derive a temporally weighted velocity-matching upper bound and discretize it into practical objectives parameterized by the number of supervision steps. Under a multi-reference setup, single-state FlowCTS-OPD outperforms vanilla KL-based OPD with faster convergence. FlowCTS-OPD improves GenEval from 0.90 to 0.93, OCR from 0.90 to 0.92, and PickScore from 22.75 to 23.06, while outperforming a mixed-reward RL baseline across all target metrics. Further analysis reveals a clear temporal supervision mismatch in vanilla KL-based OPD arising from its auxiliary SDE transition kernels. Beyond on-policy setting,FlowCTS also consistently outperforms vanilla SFT , particularly on OCR, while increasing supervision steps exhibit a trade-off between richer trajectory information and greater optimization difficulty.
Abstract:While on-policy distillation (OPD) effectively addresses sparse rewards and exposure bias in large language model post-training, its extension to flow models remains underexplored. To this end, we propose Flow Continuous Trajectory Supervision (FlowCTS), which matches subsequent student and reference trajectories initialized from the same student-visited state. Using the integral relation between trajectories and velocity fields, we derive a temporally weighted velocity-matching upper bound and discretize it into practical objectives parameterized by the number of supervision steps. Under a multi-reference setup, single-state FlowCTS-OPD outperforms vanilla KL-based OPD with faster convergence. FlowCTS-OPD improves GenEval from 0.90 to 0.93, OCR from 0.90 to 0.92, and PickScore from 22.75 to 23.06, while outperforming a mixed-reward RL baseline across all target metrics. Further analysis reveals a clear temporal supervision mismatch in vanilla KL-based OPD arising from its auxiliary SDE transition kernels. Beyond on-policy setting,FlowCTS also consistently outperforms vanilla SFT , particularly on OCR, while increasing supervision steps exhibit a trade-off between richer trajectory information and greater optimization difficulty.
Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
arXiv:2607.24522 [cs.LG]
(or arXiv:2607.24522v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2607.24522 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 263 words
Related coverage
[AINews] not much happened today
Latent Space · Sep 18, 2026
[AINews] not much happened today
Latent Space · Sep 10, 2026
FlowAWR: Online Adaptive Flow Reinforcement via Advantage-Weighted Rectification
arXiv · Oct 12, 2026
Elite-Weighted Supervised Fine-tuning for Goal-Directed Molecular Optimization
arXiv · Sep 2, 2026