Metraipolis › archive › dc_98dd1971936e
First Make It Playable, Then Make It Good: Staged Interaction Learning for Small Dialogue-Game Agents — arXiv:2608.27672
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2608.27672
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-17T10:47:53.732Z
- SHA-256 of the archived text
409150db2574a75bb082e64b6e5841d63105f48e4c563fc6136755c265b245f5- Stable URL of this page
https://metraipolis.com/documento/dc_98dd1971936e
How to cite
First Make It Playable, Then Make It Good: Staged Interaction Learning for Small Dialogue-Game Agents — arXiv:2608.27672 — arXiv (en), https://arxiv.org/abs/2608.27672 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_98dd1971936e (Metraipolis archive dc_98dd1971936e, sha256 409150db2574a75bb082e64b6e5841d63105f48e4c563fc6136755c265b245f5)
BibTeX
@misc{dc98dd1971936e,
title = {First Make It Playable, Then Make It Good: Staged Interaction Learning for Small Dialogue-Game Agents — arXiv:2608.27672},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2608.27672},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_98dd1971936e; sha256 409150db2574a75bb082e64b6e5841d63105f48e4c563fc6136755c265b245f5},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.27672. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
We present Qwen-GuidePlay-2B, a 2B-parameter language model for dialogue-game interaction. We fine-tune Qwen3.5-2B using three steps: a) SFT on only successful game trajectories from Playpen, b) weighted turn-level SFT, and c) teacher-guided SFT. The teacher model (which is a larger model) is only used to fix formatting and evaluate examples, but does not create new gold actions. Our final model scores 57.12 clemscore and 42.68 statscore on the public Playpen validation. In the officially released challenge results, our model obtains the second-highest Playpen clemscore delta among submitted systems (which is approximately +36 over its base model). Our findings suggest that imitating full trajectories helps with playability, while turn-level and teacher-guided training usually improve decision-making and increase the overall score. Alternative procedurally heavy approaches like replay-repair and hard-example mining did not help, which suggests that small models are performant simply by using careful curation strategies rather than aggressive changes. We make available both the model and the code for reproducibility. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-31 in cs.AI). Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2608.27672 — 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 2026-10-17T10:33:42.464Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-08-31?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).