LLM Researcher

Design and build high‑quality instruction datasets and QA data synthesis pipelines for chemistry and materials science, ensuring comprehensive coverage of core chemical tasks including molecular property prediction, synthesis pathway planning, and reaction mechanism analysis. 设计并搭建高质量化学与材料科学领域的指令数据集与问答集数据合成框架，确保数据覆盖分子性质预测、合成路径规划、反应机理分析等核心化学任务。 Drive post-training of domain-specific LLMs for chemistry and materials, including Supervised Fine-Tuning (SFT), Reinforcement Learning from Human Feedback (RLHF/DPO/PPO), and chemistry-specific reward model development. 负责化学材料领域大模型的后训练，包括监督微调（SFT）、基于人类反馈的强化学习（RLHF/DPO/PPO）及化学领域特定的奖励模型训练。 Research and implement multimodal alignment techniques to unify representations and enable collaborative processing of text, molecular graphs, crystal structures, spectral images, and experimental data. 研究并实现多模态对齐技术，构建文本、分子图谱、晶体结构、光谱图像及实验数据的统一表征与协同处理框架。 Develop agent systems for chemical research, integrating chemistry tools (e.g., RDKit, DFT, MLFF interfaces) to enable autonomous scientific reasoning and experimental design. 开发化学材料领域的智能体系统，集成外部化学工具（如RDKit、DFT、MLFF接口），实现自主科研推理与实验设计。 Track cutting-edge AI for Science developments and explore innovative applications of multimodal LLMs in materials discovery and battery design. 跟踪AI for Science前沿进展，探索多模态大模型在材料发现和电池设计等场景的创新应用。

Master's degree or above in Computer Science, AI, Computational Chemistry, Materials Science, or related fields, with solid theoretical foundations in machine learning and deep learning. 计算机科学、人工智能、计算化学、材料科学或相关专业硕士及以上学历，具备扎实的机器学习和深度学习理论基础。 Expertise in LLM post-training techniques with hands-on experience in SFT, RLHF, DPO algorithms; familiarity with distributed training frameworks such as DeepSpeed and Megatron-LM. 精通大模型后训练技术，具备SFT、RLHF、DPO等算法的实际落地经验，熟悉DeepSpeed、Megatron-LM等分布式训练框架。 Proficiency in Python and mainstream deep learning frameworks (PyTorch/TensorFlow), with experience in large-scale data processing and multimodal model development. 熟练掌握Python及主流深度学习框架（PyTorch/TensorFlow），具备大规模数据处理和多模态模型开发经验。 Deep understanding of Transformer architectures and attention mechanisms; familiarity with mainstream LLM architectures (LLaMA, Qwen, DeepSeek) and their training pipelines. 深入理解Transformer架构及注意力机制，熟悉LLaMA、Qwen、DeepSeek等主流大模型架构及其训练流程 Preferred Qualifications： Publication record with first-author publications in top-tier venues (e.g., NeurIPS, ICML, ICLR, etc.). 以第一作者在NeurIPS、ICML、ICLR等顶级会议或期刊发表过相关论文。 Strong domain knowledge in chemistry and materials science, including molecular representations (e.g., SMILES), and basic understanding of chemistry tools (e.g., molecular dynamics simulations). 具备良好的化学材料领域知识，熟悉SMILES等分子表示方法，了解化学计算常用工具。 Practical experience with multimodal LLMs (e.g., CLIP, BLIP or scientific LLMs (e.g., Intern-S1, ChemLLM, NatureLM). 具备多模态大模型（如CLIP、BLIP）或科学大模型（如Intern-S1, ChemLLM, NatureLM）的实际项目经验