Fine-Tuning DeepSeek-OCR-2 for Molecular Structure Recognition

Optical chemical structure recognition (OCSR) is critical for converting 2D molecular diagrams from printed literature into machine-readable formats. While vision-language models have shown promise in end-to-end OCR tasks, their direct application to OCSR remains challenging, and direct full-parameter supervised fine-tuning often fails. In this work, we adapt DeepSeek-OCR-2 for molecular optical recognition by formulating the task as an image-conditioned SMILES generation. To overcome training instabilities, we propose a two-stage progressive supervised fine-tuning strategy: starting with parameter-efficient LoRA and transitioning to selective full-parameter fine-tuning with split learning rates. We train our model on a large-scale corpus combining synthetic renderings from PubChem and realistic patent images from USPTO-MOL to improve the coverage and robustness. Our fine-tuned model, MolSeek-OCR, demonstrates competitive capabilities, achieving exact-match accuracies comparable to those of the best-performing image-to-sequence model. However, it remains inferior to state-of-the-art image-to-graph models. Exploratory post-training shows mixed results: ReFT provides small benchmark-dependent gains, whereas GSPO does not yield stable improvement and can collapse under extended optimization.