samuelfaj/distill-expert-535k

Distill Expert 535k是一个用于训练distill-expert模型的数据集，该模型是基于Qwen3-0.6B的0.6B LoRA微调模型，旨在压缩shell或命令输出以便于AI系统消费。数据集包含535,000个示例，分为训练（85%）、验证（10%）和测试（5%）三部分，覆盖8种操作模式，包括提取结构化JSON、总结长输出、分类、过滤行、二元决策（YES/NO）、通过/失败裁决、安全评估（SAFE/REVIEW/UNSAFE）以及格式转换（如JSON到YAML）。每个示例以JSON对象格式存储，包含消息（系统、用户、助手角色）、操作模式和种子ID。Distill Expert 535k is a training dataset for the distill-expert model — a 0.6B LoRA fine-tuned from Qwen3-0.6B that compresses shell/command output for AI consumption. It contains 535,000 examples split into training (85%), validation (10%), and test (5%) sets, across 8 operation modes: extract structured JSON, summarize long outputs concisely, classify output into categories, filter lines matching criteria, binary YES/NO decisions, PASS/FAIL verdicts, SAFE/REVIEW/UNSAFE assessment, and format conversions (e.g., JSON to YAML). Each record is a JSON object with messages (system/user/assistant roles), mode, and seed_id.