AnalysisAI ModelsAugust 31, 2026

Qwen-GuidePlay-2B fine-tuned for dialogue-game play

Researchers fine-tuned Qwen3.5-2B into Qwen-GuidePlay-2B using three stages: SFT on successful Playpen game trajectories, weighted turn-level SFT, then teacher-guided SFT. A companion recipe reports gains that stay largely within the targeted game family.

2 sources

More stories today

Open the live feed