mlsanityData to 13 Oct 2026

First Make It Playable, Then Make It Good: Staged Interaction Learning for Small Dialogue-Game Agents

arXiv 2608.27672 ↗First posted 27 Aug 2026cs.CL cs.AIcited by 0

On the map: LLM agents and tool use → Reinforcement learning for LLM agents

Abstract (arXiv)

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.

Key facts. No citing papers in the map yet (data through 13 Oct 2026). Placed in Reinforcement learning for LLM agents, one of 16 directions in LLM agents and tool use. Builds on 7 of its 22 references that are in the map. First posted on arXiv 27 Aug 2026.

Builds on

7 of 22 references.

Closest ancestors · 2

Same direction, newest first

  1. 2024clembench-2024: A Challenging, Dynamic, Complementary, Multilingual Benchmark and Underlying Flexible Framework for LLMs as Multi-Action Agents31 May
  2. 2023Clembench: Using Game Play to Evaluate Chat-Optimized Language Models as Conversational Agents22 May

Other references · 5

  1. 2023QLoRA: Efficient Finetuning of Quantized LLMscited by 3,388
  2. 2019Advantage-Weighted Regression: Simple and Scalable Off-Policy Reinforcement Learningcited by 767
  3. 2023RAFT: Reward rAnked FineTuning for Generative Foundation Model Alignmentcited by 680
  4. 2023Reinforced Self-Training (ReST) for Language Modelingcited by 492
  5. 2025A Third Paradigm for LLM Evaluation: Dialogue Game-Based Evaluation using clembenchcited by 3

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025Playpen: An Environment for Exploring Learning Through Conversational Interaction
  2. 2025PRINCIPLES: Synthetic Strategy Memory for Proactive Dialogue Agents
  3. 2025RoleMRC: A Fine-Grained Composite Benchmark for Role-Playing and Instruction-Following
  4. 2024How Many Parameters Does it Take to Change a Light Bulb? Evaluating Performance in Self-Play of Conversational Games as a Function of Model Characteristics
  5. 2025SpeakRL: Synergizing Reasoning, Speaking, and Acting in Language Models with Reinforcement Learning
  6. 2025Beyond Single-Turn: A Survey on Multi-Turn Interactions with Large Language Models
  7. 2025SpeechRole: A Large-Scale Dataset and Benchmark for Evaluating Speech Role-Playing Agents
  8. 2026Up and Down the Abstraction Ladder: Code-Based Skills for Language Agents