Add comprehensive documentation for MindGYM logic implementation in codebase

[Copilot]

User requested location of MindGYM (NeurIPS'25) implementation - a question synthesis system for thinking-centric fine-tuning that achieved 16% improvement on MathVision using only 400 samples.

Documentation Added

Created MindGYM_LOGIC_DOCUMENTATION.md (273 lines) mapping the complete implementation:

Core Components

• Difficulty Scoring: data_juicer/ops/filter/llm_difficulty_score_filter.py - Multi-dimensional difficulty evaluation (linguistic complexity, conceptual depth, prior knowledge, step complexity, ambiguity) on 1-5 scale
• QA Generation from Examples: data_juicer/ops/mapper/generate_qa_from_examples_mapper.py - Self-synthesis with ROUGE-L deduplication
• QA Generation from Text: data_juicer/ops/mapper/generate_qa_from_text_mapper.py - Text-to-QA conversion
• Query Optimization: data_juicer/ops/mapper/optimize_query_mapper.py - Question refinement
• Query Calibration: data_juicer/ops/mapper/calibrate_query_mapper.py - Reference-based question adjustment

Typical Pipeline

process:
  - generate_qa_from_examples_mapper:
      hf_model: 'Qwen/Qwen2.5-7B-Instruct'
      seed_file: 'seeds.jsonl'
      similarity_threshold: 0.7
  
  - llm_difficulty_score_filter:
      api_or_hf_model: 'qwen2.5-72b-instruct'
      min_score: 0.6  # Keep moderate to high difficulty
      max_score: 1.0

Documentation includes file locations, usage examples, configuration reference, workflow diagrams, and test file mappings.

Original prompt

帮我找找MindGYM的逻辑在代码哪里？

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.