metadata

dataset_info:
  features:
    - name: conversations
      list:
        - name: from
          dtype: string
        - name: value
          dtype: string
  splits:
    - name: train
      num_bytes: 79860701
      num_examples: 30000
  download_size: 38685452
  dataset_size: 79860701
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
task_categories:
  - question-answering
  - text-generation
language:
  - zh

Kyara: Knowledge Yielding Adaptive Retrieval Augmentation for LLM Fine-tuning

🤗 Hugging Face ｜ 🚀Github ｜ 📑 Paper ｜ 📖 English ｜ 📖 Chinese

This is a dataset for training Kyara.

We applied PersonaHub to generate 50K math problems and used Gemini-1.5-Flash to filter out data with obvious errors in calculation and reasoning. However, this dataset still contains some computational and logical errors; please use it with caution.