Tag
A personal project led to an ACL 2026 paper introducing TIME, a method training Qwen3 models to engage in short, context-triggered thinking rather than excessive reasoning. The work uses QLoRA and a four-phase curriculum, with all data and code released open-source.