Research
Trajectory2Gym proposes decision-grounded interaction synthesis for training tool-use agents
The method builds executable interactions around under-supported compositional structures and trains only on grounded decisions. Across five Qwen3 scales it improves Tau1, Tau2 and BFCL, raising the three-benchmark mean by 16.9 to 29.8 percentage points.
Every outlet that ran it 1 article
Hugging Face PapersResearchLead
Trajectory2Gym: Decision-Grounded Compositional Interaction Synthesis for Tool-Use Agents