多模态算法实习生（J107216）
Core
Conducting post-training (SFT/DPO) on open-source multimodal base models, synthesizing high-quality training data via LLM teachers, and building evaluation systems.
Role type
Intern, Multimodal Algorithm Engineer
Builds
Multimodal large models with tool-calling, RAG, and structured output capabilities
Domain
Artificial Intelligence, Multimodal Large Language Models
Deliverable
production ML models
Required skills
Python, PyTorch, Multimodal LLM post-training (SFT/LoRA/DPO), Data synthesis, Quantitative evaluation
Preferred skills
RAG/Vector retrieval, Agent & tool call, Model distillation, Inference optimization, Academic papers or open-source projects
Technologies
PyTorch, LLM-as-judge
Responsibilities
Design and synthesize long-horizon tool call trajectories; Participate in multimodal retrieval chain construction; Distill large model capabilities into smaller models; Build evaluation systems using LLM-as-judge and manual sampling.