Release of CCORE GLB-3D-Coder Dataset

A new dataset designed to facilitate the generation of 3D assets via Three.js and GLB export has been released by CCORE. The dataset, titled “ThreeJS-3D-GLB-Assets,” was developed to support the training of the CCORE GLB-3D-Coder, a 4B parameter model specialized in creating 3D assets.

Dataset Composition and Scope

According to documentation provided by the developer on Reddit, the dataset comprises 3,918 total examples spanning 40 distinct domains. These domains include a variety of categories such as characters, creatures, vehicles, laboratory equipment, musical instruments, architecture, machinery, nature, and everyday props.

The release includes a breakdown of 3,501 training examples and 417 validation examples. The developer stated that every example underwent testing and rendering, with the exported GLB files being validated prior to inclusion in the set.

The technical contents of the dataset include:

  • Complete conversation logs and reasoning traces.
  • HTML/JavaScript responses.
  • Rendered reference images for every example.
  • Static assets alongside examples featuring animation and interaction.

The dataset has been released under the MIT license, allowing for broader use in experiments, fine-tuning, and the development of 3D tools.

Model Integration and Availability

The dataset is intended to complement the CCORE GLB-3D-Coder model, which is a 4B parameter model. Both the dataset and the model weights are hosted on Hugging Face. The developer indicated that further releases are planned following this release.

While the developer describes the dataset as a resource for 3D tool development and fine-tuning, the claims regarding the dataset’s utility for specific experimental outcomes remain unverified by independent third-party benchmarks.

Community Reception

Initial feedback from the developer’s announcement on Reddit has been positive, with users expressing appreciation for the release of the dataset. However, as the release originated from a self-published post, the broader impact of the dataset on the 3D modeling and AI development sectors has yet to be independently measured.