腾讯游戏-高级研究员-大模型后训练
Core
Research and develop next-generation dialogue foundation models for intelligent NPCs in games, enabling real-time, human-like interactions with unique worldviews and character styles.
Role type
Senior Research Scientist (Large Language Model Post-Training)
Builds
Production-ready dialogue models for game NPCs integrated into Unreal engines
Domain
Gaming + Large Language Models (Post-Training, RL, Distillation)
Deliverable
production ML models
Required skills
Large-scale text-to-text model training, Distributed training engineering, Post-training techniques (SFT, GRPO, RLVR, RFT), Model distillation (OPD, multi-teacher), Data pipeline construction, Evaluation pipeline construction, End-to-end deployment
Preferred skills
Reasoning model training (test-time scaling, DAPO), Synthetic data generation, RL optimization
Technologies
Python, PyTorch, DeepSpeed, Megatron-Core, FSDP, vLLM, SGLang, Unreal Engine
Responsibilities
Develop dialogue foundation models for intelligent NPCs; Research and implement post-training techniques like SFT, RLVR, and distillation; Construct game-specific datasets and optimize data pipelines; Collaborate with engine teams to ensure model integration and stability; Design evaluation loops using synthetic data and user feedback; Translate business requirements into technical solutions and output documentation/patents.
