混元vlm算法研究员-后训练/基座能力提升方向（深圳/北京/上海）
Core
Lead post-training R&D for multimodal large language models (VLMs), focusing on SFT, RLHF, and alignment algorithms to enhance visual base model capabilities and reduce hallucinations.
Role type
Senior IC machine-learning researcher (VLM post-training)
Builds
Multimodal large language models with enhanced visual perception, reasoning, and instruction-following capabilities
Domain
Artificial Intelligence / Large Language Models / Computer Vision
Deliverable
production ML models
Required skills
Transformer architecture expertise, LLM alignment (SFT, RLHF, PPO, GRPO), reward modeling, RAG/memory mechanisms, Python, PyTorch/TensorFlow, distributed training (Megatron-LM, DeepSpeed, vLLM), hallucination detection and removal
Preferred skills
User persona modeling, recommendation systems integration, long-context/long-term memory optimization, high-impact publications (NeurIPS, ICLR, ICML, ACL, EMNLP), open-source contributions
Responsibilities
Design and execute post-training pipelines for VLMs; optimize visual perception and spatial understanding; develop strategies to reduce model hallucinations; align models for real-world user interaction scenarios including multi-turn dialogue and logical reasoning
