Qwen3-1.7B · ALFWorld · online ROSE 10+5 with a gpt-5.4-mini teacher — epoch 1 (first run)
Student: Qwen/Qwen3-1.7B. Part of a study comparing teacher-trajectory SFT, online multi-turn ROSE and on-policy distillation for
multi-turn agents.
Training
Online multi-turn ROSE: the student plays the first 10 turns in a live environment, the teacher continues the same environment for up to 5 turns, and only the teacher's turns are cross-entropy targets. 32 episodes per step, AdamW constant lr 1e-5, bf16, thinking off, 512 tokens per turn. Teacher: gpt-5.4-mini (OpenAI API, reasoning_effort=medium; only the visible reply is used). 2181 training games, 68 steps per epoch. This run was stopped at step 95; this is its epoch-1 checkpoint (step 68).
Evaluation
ALFWorld test split (AgentGym, 200 games), avg of 4 passes, temperature 0.4, 512 tokens/turn, 30 rounds, thinking off. Success 34.50% ± 2.42. The repeat run's epoch 1 scored 32.00% (difference +2.50, 95% CI [−1.00, 6.12]). Reference points on the same protocol: base Qwen3-1.7B 19.38%, Qwen3-32B itself 52.12%.
Prompt format
ReAct, multi-turn chat: the environment's AgentGym instruction as the first user turn, a canned assistant acknowledgement, then one user
turn per observation; the model replies Thought:\n...\n\nAction:\n<one action>. Render with Qwen3's chat template and enable_thinking=False.