Ricardo-H/sciworld-global2-3b-to-7b

GATS student — ScienceWorld, Qwen2.5-3B-Instruct → Qwen2.5-7B-Instruct

Gap-Adaptive Teacher Scheduling (GATS) student checkpoint from "Guide, Then Let Go: Gap-Adaptive Teacher Scheduling for Sparse-Reward Agentic RL".

Field	Value
Environment	ScienceWorld
Teacher	Qwen/Qwen2.5-3B-Instruct (GRPO-trained, frozen)
Student	Qwen/Qwen2.5-7B-Instruct
Method	GATS (adaptive OPD weight, irreversible handoff)
Training steps	150

Part of the GTL collection.

License note: derived from Qwen2.5-3B-Instruct, distributed under the Qwen Research License rather than Apache-2.0. Non-commercial use only.