README.md

---
dataset_info:
  features:
  - name: system
    dtype: string
  - name: input
    dtype: string
  - name: output
    dtype: string
  - name: conversations
    list:
    - name: from
      dtype: string
    - name: value
      dtype: string
  splits:
  - name: train
    num_bytes: 4229806
    num_examples: 3530
  - name: test
    num_bytes: 36925
    num_examples: 29
  download_size: 502496
  dataset_size: 4266731
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: test
    path: data/test-*
---
# Home-Assistant-Requests-Zh

**Home-Assistant-Requests-Zh** is a dataset designed to train large language models (LLMs) in Traditional Chinese for the purpose of controlling smart home systems, specifically targeting Home Assistant systems. The dataset contains user inputs in Traditional Chinese, and the outputs are structured JSON commands that represent the user's intent to control smart home devices.

This dataset was derived from the original Simplified Chinese dataset [Charles95/smart_home_control](https://huggingface.co/datasets/Charles95/smart_home_control). The input and output fields have been converted to Traditional Chinese to make the dataset suitable for Traditional Chinese LLMs.

- **Total examples**: 3,530
- **Languages**: Traditional Chinese

## Dataset Structure

The dataset contains the following columns:

- **system**: A fixed prompt to instruct the model on how to interpret the user's input and generate a structured control command. The prompt used is: `"你是一個智慧家居助手，請你基於使用者輸入的文本，解析語義並抽取關鍵資訊，以json格式生成結構化的智慧家居控制指令。"`.
- **input**: User-provided requests in Traditional Chinese, asking to control specific smart home devices or functionalities.
- **output**: Expected structured output in JSON format, detailing the user's intent and associated parameters (device, action, attributes, etc.) for controlling the smart home system.