Metraipolis › archive › dc_bc494b3166e2

Beyond Forgetting: Diagnosing and Harnessing Shared Reasoning in Continual RLVR — arXiv:2608.18574

Source: arXiv (en)
Original: https://arxiv.org/abs/2608.18574
License (as archived): CC0-1.0
Archived: 2026-10-18T13:53:17.546Z
SHA-256 of the archived text: 08899b15a30151a2ce0ac64895af34c4e4d2f7584cbfd367125b5965fa2db270
Stable URL of this page: https://metraipolis.com/documento/dc_bc494b3166e2

How to cite

Beyond Forgetting: Diagnosing and Harnessing Shared Reasoning in Continual RLVR — arXiv:2608.18574 — arXiv (en), https://arxiv.org/abs/2608.18574 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_bc494b3166e2 (Metraipolis archive dc_bc494b3166e2, sha256 08899b15a30151a2ce0ac64895af34c4e4d2f7584cbfd367125b5965fa2db270)

BibTeX

@misc{dcbc494b3166e2,
  title = {Beyond Forgetting: Diagnosing and Harnessing Shared Reasoning in Continual RLVR — arXiv:2608.18574},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.18574},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_bc494b3166e2; sha256 08899b15a30151a2ce0ac64895af34c4e4d2f7584cbfd367125b5965fa2db270},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.18574. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Reinforcement learning with verifiable rewards (RLVR) commonly post-trains reasoning models on multiple tasks, while rerunning multitask RLVR (MTRL) as new tasks are added makes capability expansion costly. We therefore study continual RLVR, which updates the existing model as each task arrives. The central question is whether a model updated this way can perform as well as a jointly trained model. To answer this question, we introduce Continual Reasoning Gym, a continual-RLVR environment that organizes text and visual reasoning tasks into five task sequences. In this setting, we identify two key observations: Sequential RLVR exhibits modest forgetting, yet its final performance remains below that of MTRL. To understand the latter, we decompose final performance and show that forgetting accounts for only part of the gap. To explain the former, we identify shared reasoning: transferable reasoning structure allows training on one task to support others on average. We therefore introduce Continual Prompt Replay (CPR), which harnesses shared reasoning to improve learning on the arriving and future tasks by replaying previous-task prompts and regenerating their responses with the current policy. On average, only CPR reaches MTRL-level performance.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-20 in cs.LG).
Categorie arXiv: Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2608.18574 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-18T13:50:58.914Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-08-20?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).