Research

Research behind Spectra

ROSE is ROMBO AI's public 1H NMR foundation model. Spectra is the commercial identification product.

Two public objects should stay distinct.

Spectra, the product

Spectra is the commercial web application for NMR compound identification. It retrieves candidates, ranks them, and compares predicted and measured shifts.

ROSE, the research model

ROSE (Reusable One-dimensional Spectrum Embeddings) is a public 1H NMR foundation model released by ROMBO AI. The published materials describe a 7.8M-parameter encoder trained on the order of 3.2 million 1H spectra, with weights on Hugging Face and code on GitHub.

ROSE is intended for research and prototyping on 1D 1H spectra: embeddings, denoising, peak maps, retrieval heads and related tasks. ROSE is not a structure-elucidation solver.

Fact	Value	Source
Parameters	7.8M	Hugging Face model card
Pretraining spectra	about 3.2 million 1H	ROSE repository
Preprint	ChemRxiv	10.26434/chemrxiv.15007823/v1
Archive	Zenodo	10.5281/zenodo.22142631

Links

• ROSE GitHub repository
• ROSE weights on Hugging Face
• ChemRxiv preprint
• Zenodo archive
• ROMBO AI
• The missing layer in NMR machine learning

Cite Spectra

Spectra by ROMBO AI. AI-assisted NMR compound identification. https://spectra.rombo.ai

Cite ROSE

Use the BibTeX published with the model:

@article{diiorio2026rose,
  title   = {ROSE: a Foundation Model for Reusable One-dimensional
             Spectrum Embeddings in 1H NMR},
  author  = {Di Iorio, Mattia and Mattia, Carmine and Zanda, Andrea
             and Atzori, Maurizio},
  year    = {2026},
  journal = {ChemRxiv},
  doi     = {10.26434/chemrxiv.15007823/v1},
  url     = {https://doi.org/10.26434/chemrxiv.15007823/v1},
  note    = {Preprint}
}

Try Read the docs