Import from Hugging Face

• Access to Diverse Datasets: Leverage high-quality, curated datasets for training and testing.
• Preprocessed Data: Many datasets are formatted and ready to use.
• Easy Integration: Directly imports into the platform without manual conversion.

This feature streamlines the process of working with established datasets, making it faster and more efficient to get started with data-driven experiments.

​

Steps to Import a Hugging Face Dataset

​

1. Access the Dataset Import Section

• Navigate to the Datasets & Experiments section.

• Click on “Add Dataset” to access dataset creation options.

• Select “Import from Hugging Face” from the available choices.

​

2. Browse and Select a Dataset

• The system presents a catalog of datasets sourced from Hugging Face.
• Each dataset includes key metadata such as:
  • Dataset Name (e.g., databricks-dolly-15k)
  • Source (e.g., OpenAI, Hugging Face, Microsoft)
  • Record Count
  • Usage Popularity and Metadata
• Use the search functionality to locate a specific dataset.

​

3. Configure Dataset Parameters

• Upon selection, a configuration panel appears displaying:

  • Dataset Overview: Summary, source, and dataset reference link.

  • Subset Selection: Options include Default, Train, or Split.

  • Number of Rows: Specify the number of records to be imported.

  • Additional Preferences: Optionally enable “Add selected rows” for precise filtering.

​

4. Initiate the Import Process

• Click “Start Experimenting” to commence the dataset ingestion.
• The imported dataset will be available in the Datasets & Experiments section for further processing and utilization.