LASR-Callum/

2026-08-04-qwen36-table2-80-synthdoc-self-reflect-20-sft-bundle

	Qwen3.6 Table2 80% + SynthDoc self-reflection 20% training bundle




	

field value

experiment One-epoch Qwen3.6-27B assistant-only LoRA SFT, mixing filtered Table-2 instruction data with first-person SynthDoc self-reflection at 80/20 by loss-bearing tokens.

date_generated 2026-08-04

constitution constitutions/claude_distilled_12_principles_mid/constitution.md; both upstream corpora connect to this target.

source_repo Matthew-Bozoukov/teaching_claude_why_replication… See the full description on the dataset page: https://huggingface.co/datasets/LASR-Callum/2026-08-04-qwen36-table2-80-synthdoc-self-reflect-20-sft-bundle.