LLM & Multimodal Post-Training Research Scientist

Alibaba Group; Sunnyvale (CA); Presencial; USD 120 000 - 180 000; Tempo integral

Resumo da oferta

Alibaba Group in Sunnyvale, CA is seeking outstanding research scientists and engineers for the next generation of LLM and Multimodal Agent systems. You will advance post-training methods and agent capabilities to enable complex tasks such as DeepResearch, tool use, coding, and long-horizon reasoning.

You will collaborate with leading researchers and engineers, design training approaches like Agentic SFT and CPT, build large-scale datasets, and contribute to open-source projects or publications

Qualificações

• Master’s or PhD in CS/AI or related field.
• Experience with LLM training or post-training (fine-tuning).
• Publications in NeurIPS, ICML, ICLR or CVPR, or significant open-source contributions.

Responsabilidades

• Research and develop post-training methods for LLM and multimodal models.
• Design and implement training approaches such as Agentic SFT, Agentic CPT, Reasoning RL, and Agentic RL.
• Build large-scale agent training datasets for complex tasks like DeepResearch, tool use, and coding.
• Design and implement agent evaluation frameworks for complex task execution and tool-use scenarios.
• Collaborate with model training, data engineering, and systems teams to translate research into real-world AI systems and products.
• Contribute to the AI community through open-source projects or publications.

Conhecimentos

LLM training

Multimodal systems

Deep research

Formação académica

Master’s or PhD in CS/AI

Descrição da oferta de emprego

Alibaba Group in Sunnyvale, CA is seeking outstanding research scientists and engineers for the next generation of LLM and Multimodal Agent systems. You will advance post-training methods and agent capabilities to enable complex tasks such as DeepResearch, tool use, coding, and long-horizon reasoning.

You will collaborate with leading researchers and engineers, design training approaches like Agentic SFT and CPT, build large-scale datasets, and contribute to open-source projects or publications