MolDeTr — 1D Deformable-DETR for automated ¹H NMR multiplet detection

Trained weights for MolDeTr, a chemistry-informed deep-learning model that detects the multiplets in a 1-D ¹H NMR window and returns each spin system's chemical shift (δ), coupling (J), proton count, and line width in a single forward pass.

• Code: https://github.com/smidooo/MolDeTr
• Paper: Analytical Chemistry 2026 — 10.1021/acs.analchem.5c03465
• Data + weights (Zenodo): 10.5281/zenodo.21217102

Files

• model_spin_system_ABCDEFG_exp2.pth (~974 MB) — the trained checkpoint.

Usage

from huggingface_hub import hf_hub_download
ckpt = hf_hub_download("smidooo/moldetr", "model_spin_system_ABCDEFG_exp2.pth")
# then point MolDeTr at it, e.g. set MOLDETR_CHECKPOINT=ckpt (see the repo README).

CPU inference works out of the box (the deformable-attention op has a pure-PyTorch fallback). License: Apache-2.0.