Metraipolis › archive › dc_e2a8b5364cb5
CounterRoute: Self-Routed Reasoning via Hierarchical Counterfactual Credit Assignment — arXiv:2609.29109
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.29109
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-08T04:25:54.332Z
- SHA-256 of the archived text
cda3d9b314847db62cfa6096fca2461e32e9bdb8c2cc969b642b4de15e38baab- Stable URL of this page
https://metraipolis.com/documento/dc_e2a8b5364cb5
How to cite
CounterRoute: Self-Routed Reasoning via Hierarchical Counterfactual Credit Assignment — arXiv:2609.29109 — arXiv (en), https://arxiv.org/abs/2609.29109 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_e2a8b5364cb5 (Metraipolis archive dc_e2a8b5364cb5, sha256 cda3d9b314847db62cfa6096fca2461e32e9bdb8c2cc969b642b4de15e38baab)
BibTeX
@misc{dce2a8b5364cb5,
title = {CounterRoute: Self-Routed Reasoning via Hierarchical Counterfactual Credit Assignment — arXiv:2609.29109},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.29109},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_e2a8b5364cb5; sha256 cda3d9b314847db62cfa6096fca2461e32e9bdb8c2cc969b642b4de15e38baab},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.29109. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Reasoning-capable language models often produce long chains of thought when direct answers suffice, wasting inference compute. Dual-mode models offer both thinking and direct-answer modes, but typically leave mode selection to users. Automating this selection while improving responses under both modes is challenging because routing targets evolve with the policy, strong initial mode preferences can destabilize exploration, and sequence-level objectives entangle routing with response learning. We introduce CounterRoute, an online reinforcement-learning framework that jointly learns routing and mode-conditioned responses in one shared policy directly from a native dual-mode checkpoint, without method-specific SFT warm-up. Paired current-policy counterfactual rollouts estimate cross-mode routing credit, which is assigned only to the routing token, while within-mode GRPO trains response tokens. A paired-to-self-routed curriculum stabilizes early training with forced coverage of both modes before progressively transferring training to autonomous routing. Across nine benchmarks, CounterRoute better balances accuracy and efficiency than heuristic and learned adaptive-routing methods, improving macro-average accuracy at both Qwen3 scales while cutting mean generated tokens by 51% and 41%, respectively, versus always-thinking checkpoints. On instruction-following and commonsense benchmarks where direct answering is strong, think rates fall as low as 1% while response quality improves. Despite training only on math and synthetic instruction-following data, its routing and response quality generalize to held-out tasks in coding, science, knowledge, and commonsense. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-25 in cs.AI). Categorie arXiv: Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2609.29109 — 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 2026-10-08T03:50:57.716Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-25?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).