Fine-Tuning DeepSeek-OCR‑2 for Molecular Structure Recognition
Haocheng Tang, Xingyu Dang, Junmei Wang · 2026
Read Fine-Tuning DeepSeek-OCR‑2 for Molecular Structure Recognition by Haocheng Tang, Xingyu Dang, Junmei Wang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-13629064.txt.
Description
Tang, Dang, and Wang tune DeepSeek-OCR-2 to read chem forms from print shots. Form reads turn drawn forms in books to code text for machines. Sight-word tools show hope for end-to-end reads in such tasks. Yet chem reads strain them, and full tune oft fails cold. Their fix casts the task as shot-led SMILES text birth. A two-stage tune plan steadies the train runs in the work.
Stage one uses low-rank adds, a lean tune form, to start. Stage two swaps to picked full tune with split pace marks. Train sets join fake PubChem draws with real sheet shots. The tuned tool takes the name MolSeek-OCR in the text. Its true-match scores tie the best chain-form peer in runs. Graph-form tools still top it on those same runs.
Catalog record
| CREATOR | Haocheng Tang, Xingyu Dang, Junmei Wang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 25 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |