{
  "name": "ru-en-humanlike-synthetic-instructions",
  "version": "2.0",
  "created_at": "2026-07-16T13:07:46.652723Z",
  "rows": 20000,
  "splits": {
    "train": 19600,
    "validation": 200,
    "test": 200
  },
  "languages": [
    "ru",
    "en"
  ],
  "columns": [
    "id",
    "lang",
    "category",
    "prompt",
    "response",
    "messages",
    "source",
    "license",
    "quality_note"
  ],
  "categories": [
    "advice",
    "api_design",
    "brainstorm",
    "coding",
    "coding_debug",
    "coding_refactor",
    "coding_review",
    "coding_tests",
    "compare",
    "creative",
    "data_analysis",
    "email",
    "empathy",
    "explain",
    "greeting",
    "habit",
    "planning",
    "rewrite",
    "roleplay",
    "safety_refusal",
    "smalltalk",
    "summarize",
    "translate"
  ],
  "category_counts": {
    "advice": 1894,
    "api_design": 90,
    "brainstorm": 524,
    "coding": 1920,
    "coding_debug": 1536,
    "coding_refactor": 1176,
    "coding_review": 1178,
    "coding_tests": 1153,
    "compare": 90,
    "creative": 691,
    "data_analysis": 90,
    "email": 90,
    "empathy": 2101,
    "explain": 2448,
    "greeting": 2006,
    "habit": 90,
    "planning": 2457,
    "rewrite": 90,
    "roleplay": 90,
    "safety_refusal": 16,
    "smalltalk": 90,
    "summarize": 90,
    "translate": 90
  },
  "license": "CC0-1.0 for generated template text",
  "warning": "Synthetic template data improves format/style coverage but should be mixed with reviewed human/LLM data and deduplicated for high-quality model training."
}