研究人员开发了Bolek，一个紧凑型多模态语言模型，专为分子推理而设计。该模型将分子结构嵌入整合到指令微调的文本解码器中，使其能够将自然语言解释与具体的分子特征联系起来。在各种分子对齐和推理任务上，Bolek的表现优于Qwen3-4B-Instruct和TxGemma-9B-Chat等现有模型，能够生成更具可审计性和更可靠的解释。 AI
影响 引入了一个更具可审计性和更紧凑的分子推理模型，有望改善药物发现的决策过程。
排序理由 该集群描述了一篇关于一种新颖的用于分子推理的多模态语言模型的研究论文。
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
arXiv:2605.02745v1 Announce Type: new Abstract: Molecular property models increasingly support high-stakes drug-discovery decisions, but their outputs are often difficult to audit: classical predictors return scores without rationale, while language models can produce fluent expl…
Molecular property models increasingly support high-stakes drug-discovery decisions, but their outputs are often difficult to audit: classical predictors return scores without rationale, while language models can produce fluent explanations weakly grounded in the input molecule. …