{ "base_model": "Qwen/Qwen3-8B", "binding_mode": "environment", "n_train_rows": 700, "subagents": [ "extractor", "reasoner", "rule_applier" ], "manager_adapter": "outputs/manager/deepseek_legal_L_8B_8n/sft_evolved", "full_parameter_rl": true, "subagent_server_url": "http://localhost:8000", "routing_efficiency_bonus": 0.0, "tool_use_bonus": 1.0 }