Loom_01 is a curated dataset of 50,000 interleaved tutorials for training and evaluating diffusion-transformer models. It was created by researchers from Beijing Institute of Technology, Alibaba Group, and the National University of Singapore and released on Hugging Face in 2026.
Use Cases
- Training diffusion-transformer models based on the described interleaved tutorial data.
- Benchmarking model performance on interleaved generation tasks using the curated dataset.
- Exploring the capabilities of multimodal or sequential generation models based on the dataset's interleaved nature.
Strengths
- Contains 50,000 curated interleaved tutorial examples.
- Created by a research team from multiple academic and industry institutions.
Limitations
- Description metadata is limited; actual data quality requires manual inspection after download.
- Column-level documentation is absent; field semantics must be inferred after download.
Provenance
- Source
- Beijing Institute of Technology, Alibaba Group, National University of Singapore
- Collection Method
- Curated by researchers for the Loom project.
- Freshness
- Last updated 2026-06-02 18:21:57; freshness should be verified.