vqa / README.md
nielsr's picture
nielsr HF Staff
Add dataset card and paper/code links
3c1c7cd verified
|
Raw
History Blame
2.06 kB
metadata
license: cc-by-4.0
task_categories:
  - image-text-to-text

SupraBench: A Benchmark for Supramolecular Chemistry

SupraBench is the first benchmark for evaluating large language models on supramolecular host–guest chemistry reasoning. It comprises four fundamental tasks plus an auxiliary vision task, and provides a domain text corpus (SupraPMC) for domain-adaptive pretraining (DAPT).

Tasks & Datasets

Dataset Task Description
bap Binding Affinity Prediction Regress log $K_a$ for a host–guest pair
tbs Top-Binder Selection Pick the strongest binder among 4 candidate guests
sid Solvent Identification 6-way solvent classification from structure
hgd Host-Guest Description Open-ended QA on host/guest property profiles
vqa Molecular Identification Auxiliary vision task: identify a molecule from its image
EU-PMC Text corpus ~16M-token supramolecular corpus for DAPT
Binding-Affinity Comprehensive anchor Per-record binding data + host/guest SMILES, 2D, 3D, environment

Sample Usage

The benchmark uses uv for dependency management. You can run a task against a model using the following command found in the repository:

uv run python src/main.py \
    --task-config  configs/tasks/bap_base.yaml \
    --model-config configs/models/openrouter_qwen35_27b.yaml \
    --output-dir   outputs/

Citation

@article{ma2026suprabench,
  title   = {SupraBench: A Benchmark for Supramolecular Host--Guest Chemistry Reasoning in Large Language Models},
  author  = {Ma, Tianyi and Ma, Yijun and Wang, Zehong and Sun, Weixiang and Li, Ziming and Schmidt, Connor R. and Zhang, Chuxu and Webber, Matthew J. and Ye, Yanfang},
  year    = {2026},
  note    = {arXiv preprint, link coming soon}
}