Import from Hugging Face

Import Q&A datasets directly from Hugging Face Hub into your FloTorch workspace. This guide explains what you need and how to use the import form.

Prerequisites

• A dataset created with HuggingFace as the source (in Dataset Form → Source)
• The dataset repository ID from Hugging Face
• Knowledge of the file structure in the target dataset (file names and column names)

Where to Import

1. Open your Workspace → Datasets
2. Open a dataset with HuggingFace source
3. Go to the Files tab → click Add → Import from Hugging Face
4. Or: Dataset detail page → Add → Import from Hugging Face

Form Fields

Required

Field	Description	Example
Repository	Hugging Face dataset repository ID in username/dataset-name format	allenai/mmlu
GT File	Filename of the ground truth / evaluation file in the repo	test.jsonl, validation.jsonl
Question Map From	Name of the source column that contains the question/prompt	input, question, prompt
Answer Map From	Name of the source column that contains the expected answer	output, answer, target
Limit	Number of Q&A pairs to import	100

Optional

Field	Description	Example
Access Token	Hugging Face token for private repositories. Leave empty for public datasets	hf_xxxx...
Examples File	Filename for few-shot examples (used for N-Shot Prompts)	train.jsonl
Examples Question Map From	Column name for questions in the examples file	input
Examples Answer Map From	Column name for answers in the examples file	output

How to Find Values

1. Repository ID

• Browse datasets: huggingface.co/datasets
• From the dataset page URL: https://huggingface.co/datasets/allenai/mmlu → use allenai/mmlu

2. File Names and Columns

• Open the dataset page on Hugging Face
• Go to the Files and versions tab to see available files (e.g. train.jsonl, test.jsonl)
• Use the Preview or Explore feature to inspect the structure and column names

3. Access Token (Private Repos Only)

• Go to Hugging Face → Settings → Access Tokens
• Create a token with read access
• Paste it in the Access Token field when importing a private dataset

Example: allenai/mmlu

Field	Value
Repository	allenai/mmlu
GT File	test.jsonl (or the appropriate split)
Question Map From	input
Answer Map From	target
Limit	100

After Import

• An import job runs in the background
• Progress is shown in the dataset Files tab
• When complete, the imported Q&A pairs appear as ground truth (and optionally examples) data