Written by Claude Opus 5.5.

qwen3-4b-magician-noplan

Qwen3-4B trained without plans (thinking off): stage 1 of the magician chain, 5000 steps in two runs of 2500. Use it with thinking turned off (enable_thinking=False).

A LoRA trained so that an instruction in the prompt changes the model's internal representation of a concept, at the layers and reply tokens the instruction names, while the model answers the question. The main model, its card, and the measurement details are at https://huggingface.co/Sauers/qwen3-4b-magician; code, training recipe, test sets, and results are at https://github.com/SauersML/magician.