Qwen3-1.7B · ScienceWorld · online ROSE 10+5 with a gpt-5.4-mini teacher — separate 1-epoch run, step 32
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). 2059 training task variations, 64 steps = 1 epoch; this is step 32. A separate run from the 3-epoch series (…-ep1 … …-ep6), same hyper-parameters.
Evaluation
ScienceWorld test split (AgentGym, 200 task variations), avg of 4 passes, temperature 0.4, 512 tokens/turn, 30 rounds, thinking off, no system turn. Success: not evaluated (half an epoch). Reference points on the same protocol: base Qwen3-1.7B 0.12%, Qwen3-32B itself 15.62%.
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.