Disclosure: I created NMR Workbench, and this is a free dataset release. The spectra are synthetic; the NMRium browser interactions, screenshots and saved application states were actually recorded.<\/p>\n
NMR helps chemists study molecular structure. This collection focuses on the software steps used to correct phase and chemical-shift references.<\/p>\n
The release contains 11,270 verified workflows, 96,922 GUI actions and 108,192 screenshots\/SFT examples (approximately 75.1 GB).<\/p>\n
Demo and original announcement on X:<\/p>\n
https:\/\/x.com\/ubermensch_hb\/status\/2107894949601759371<\/a><\/p>\n Each accepted workflow was saved, reopened in a fresh browser and checked numerically against a known reference. Tasks include zero-order phase correction, chemical-shift referencing, combined corrections and already-correct controls.<\/p>\n The sft configuration pairs a screenshot, instruction and prior actions with a target action. The trajectories configuration includes before\/after observations. Parquet data and saved evidence are available directly here:<\/p>\n https:\/\/huggingface.co\/datasets\/priyanshu-harshbodhi\/nmr-workbench-large-v2<\/a><\/p>\n Scope: reference-assisted scripted demonstrations on synthetic spectra, a development split, and no established model-improvement result. This is intended for next-action training experiments, not clinical use or an independent test benchmark.<\/p>\n I’d appreciate feedback on the schema and what additional data would make it useful for scientific computer-use research.<\/p>\n<\/div>\n submitted by \/u\/Both-Performance6814 <\/a>
[link]<\/a><\/span> [comments]<\/a><\/span><\/p>
