LASR-Callum/

2026-08-14-table2-9284-difficult-advice-716-train

	1-epoch LoRA SFT mixture for Qwen3.6-27B: 9,284 filtered Table2 instruction rows + 716 trait-balanced difficult-advice-v2 rows (7.16% synthetic), for the constitution-internalisation sweep.




	

field value

experiment 1-epoch LoRA SFT mixture for Qwen3.6-27B: 9,284 filtered Table2 instruction rows + 716 trait-balanced difficult-advice-v2 rows (7.16% synthetic), for the constitution-internalisation sweep.

date_generated 2026-08-14

constitution… See the full description on the dataset page: https://huggingface.co/datasets/LASR-Callum/2026-08-14-table2-9284-difficult-advice-716-train.