video1.0<iframe src="https://www.loom.com/embed/0f624c26551b4273b33371358c3164d3" frameborder="0" width="1668" height="1251" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>12511668Loomhttps://www.loom.com12511668https://cdn.loom.com/sessions/thumbnails/0f624c26551b4273b33371358c3164d3-6719ec66493fe0f9.gif316.24In this video, I discuss my recent work accepted by ICLR 2026, which focuses on bridging large-language models with molecular graphs. We address the challenges posed by different molecular modalities and propose two key innovations: entropy-guided patching to preserve molecular structure and a DynamicQueryFormer that enhances the Q-former framework. Our model demonstrates superior performance across various benchmarks, including multiple-choice questions and property generation tasks. I encourage viewers to review our findings and consider the implications for future research in this area.