2026-08-20-difficult-advice-t10-curiosity-smoke
收藏资源简介:
该数据集是“困难建议(difficult_advice)”合成数据生成流程的阶段性快照集合,用于构建训练数据。数据基于宪法(constitution)文件(trait10_curiosity)生成,包含多个预处理与生成阶段:从宪法分块、场景编写、场景去重、提示草稿、提示修订、回复草稿、回复修订,到最终导出为SFT(监督微调)格式。每个阶段都有对应的JSONL文件作为快照,部分阶段还包含中间暂存版本(.partial)。数据集标签表明其为合成训练数据,适用于困难建议场景,且基于好奇心特质。数据生成日期为20260820_120319,来源仓库为GitHub上的特定提交。该数据集可用于训练模型在复杂或困难建议场景下的生成能力,尤其适用于需要表现出好奇心特质的对话或建议任务。
This dataset is a collection of phase snapshots from the synthetic data generation pipeline for difficult advice, used to construct training data. The data is generated based on the constitution file (trait10_curiosity) and includes multiple preprocessing and generation stages: from constitution chunking, scenario writing, scenario deduplication, prompt drafting, prompt revision, response drafting, response revision, to final export in SFT format. Each stage has a corresponding JSONL file as a snapshot, with some stages also including intermediate partial versions. The dataset labels indicate it is synthetic training data suitable for difficult advice scenarios and based on the curiosity trait. The data generation date is 20260820_120319, and the source repository is a specific commit on GitHub. This dataset can be used to train models in generating responses in complex or difficult advice scenarios, particularly suitable for dialogue or advice tasks that require exhibiting curiosity traits.
数据集详情
基本信息
- 数据集名称:synth
difficult_advicerun — per-stage snapshots (resumable generation cache) - 生成日期:20260820_120319
- 数据集地址:https://huggingface.co/datasets/LASR-Callum/2026-08-20-difficult-advice-t10-curiosity-smoke
数据集属性
| 属性 | 值 |
|---|---|
| 标签 | training-data, kind:synth, pipeline:difficult-advice, constitution:trait10_curiosity, smoke, backfilled-2026-08-25 |
| 类型 | 合成训练数据(synth) |
| 流水线 | difficult-advice |
| 宪法 | scratch/trait10_curiosity/constitution.md(特质10:好奇心) |
| 来源仓库 | https://github.com/Matthew-Bozoukov/Lessons_from_constituitional_AFT.git @ 432c0693067910a134add164588e51b3a75e1998 |
| 来源命令 | uv run synth run --config configs/data/synth/difficult_advice.yaml |
数据配置
数据集包含一个主配置 dataset(数据文件为 dataset.jsonl)以及多个阶段快照配置,每个配置对应一个独立的 JSONL 文件:
| 配置名称 | 数据文件 |
|---|---|
dataset |
dataset.jsonl |
stage_1_chunk_constitution |
stages/stage_1_chunk_constitution.jsonl |
stage_2_write_scenarios |
stages/stage_2_write_scenarios.jsonl |
stage_3_dedupe_scenarios |
stages/stage_3_dedupe_scenarios.jsonl |
stage_4_draft_prompts |
stages/stage_4_draft_prompts.jsonl |
stage_5_revise_prompts.partial |
stages/stage_5_revise_prompts.partial.jsonl |
stage_5_revise_prompts |
stages/stage_5_revise_prompts.jsonl |
stage_6_draft_responses |
stages/stage_6_draft_responses.jsonl |
stage_6_draft_responses.partial |
stages/stage_6_draft_responses.partial.jsonl |
stage_7_revise_responses |
stages/stage_7_revise_responses.jsonl |
stage_7_revise_responses.partial |
stages/stage_7_revise_responses.partial.jsonl |
stage_8_export_sft |
stages/stage_8_export_sft.jsonl |
数据结构与用途
- 架构:包含各阶段的
stage_<n>_<name>.jsonl快照文件以及manifest.json清单文件。 - 用途:该数据集为可恢复的生成缓存,保存了合成数据生成流水线各阶段的中间结果。
manifest.json中记录了完整的运行配置、采样设置、各阶段使用的模型及使用情况。 - 数据来源:通过
difficult_advice合成数据流水线生成,基于特质10(好奇心)的宪法文档进行约束。




