Error code:   ConfigNamesError
Exception:    RuntimeError
Message:      Dataset scripts are no longer supported, but found Training-dataset.py
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/dataset/config_names.py", line 67, in compute_config_names_response
                  config_names = get_dataset_config_names(
                      path=dataset,
                      token=hf_token,
                  )
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 161, in get_dataset_config_names
                  dataset_module = dataset_module_factory(
                      path,
                  ...<4 lines>...
                      **download_kwargs,
                  )
                File "/usr/local/lib/python3.14/site-packages/datasets/load.py", line 1215, in dataset_module_factory
                  raise e1 from None
                File "/usr/local/lib/python3.14/site-packages/datasets/load.py", line 1175, in dataset_module_factory
                  raise RuntimeError(f"Dataset scripts are no longer supported, but found {filename}")
              RuntimeError: Dataset scripts are no longer supported, but found Training-dataset.py

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

Lumina Training Dataset 🦋

This is the official dataset collected and released by Butterfly AI Labs under the open Apache 2.0 license. It contains exactly 100,000 high-quality "image + text" pairs taken from the well-known MS COCO 2017 database. This dataset was created specifically for training complex neural networks that generate images.

🧠 Dataset Overview

Unlike the huge datasets of large corporations, which are full of "junk" from the internet and watermarks, this dataset was collected by an independent solo developer. The creator's main idea: to make the text match the image as accurately as possible, while using less computing power.

Total pairs: 100,000 (each image has one corresponding text file)
Image size: All images have already been cropped and resized to a uniform 512x512 pixels
Text format: Strict 1-to-1 correspondence (1 text description file per 1 image)
Data source: These are specially selected images from the MS COCO 2017 database. They were carefully checked to ensure correct object shapes, varied lighting, and a clear separation between the main subject and the background

📁 Folder Structure (Repository)

All files are laid out very simply and neatly, so that programs in PyTorch or JAX can automatically read them:

Training-dataset-for-Lumina/ — main folder
├── images/ — folder with images (e.g., 000001.jpg, 000002.jpg)
├── texts/ — folder with text descriptions for these images (e.g., 000001.txt, 000002.txt)
└── Training-dataset.py — ready-made script for automatically downloading the data from the Hugging Face website

🛠️ How to Use with Streaming (On-the-fly Loading)

You don't need to download dozens of gigabytes of data to your computer. You can use the datasets library from Hugging Face to load data on the fly during neural network training.

Below in the file is an example of Python code that shows how to connect this dataset with a single command, run it, and loop through it to retrieve images and texts one at a time.

📜 License and Attribution

This dataset is distributed under the open Apache License 2.0 (it can be used even in commercial projects). If you use this data or the Lumina model architecture in your research or products, the authors ask that you credit them (cite the Butterfly AI Labs project, 2026).

Downloads last month: 67

Total file size:

11.5 GB

Models trained or fine-tuned on butterfly-ai-labs/Training-dataset

butterfly-ai-labs/Lumina_1.0_Deprecated

Text-to-Image • Updated Sep 5 • 1 • 1