大模型应用算法工程师-剪映CapCut
Core
Develop preference alignment algorithms for CapCut's multimodal models to enable domain knowledge modeling for video creation using SFT, RLHF, and POST Training techniques.
Role type
Senior IC large model application algorithm engineer (video creation)
Builds
Intelligent video creation expert agents with enhanced safety and instruction-following capabilities
Domain
Video creation, Large Language Models (LLM), Multimodal AI
Deliverable
production ML models
Required skills
SFT/RLHF/POST Training, Reinforcement Learning (PPO), Python, C++, PyTorch, Tensorflow, Deepspeed, Megatron, vLLM
Preferred skills
RLHF engineering practice, Multimodal alignment, Step-wise supervised learning
Responsibilities
Model fine-tuning for video creation domain, Enhancing pre-trained model capabilities for video tasks, Building industry-leading video creation agents