Job Description
About Us
Sea Group is establishing a brand-new, strategic AI department. This department is dedicated to exploring the transformative potential of generative AI in revolutionizing human connection, self-expression and communication diversity, and social interaction. We are building the next generation of AI-native applications and a comprehensive Model-as-a-Service (MaaS) product support system. Based on massive multi-country data, we are building a leading multilingual AI ecosystem from the ground up. We look forward to more outstanding talents joining us to build leading Southeast Asian multilingual models and explore innovative AI-native applications.
The team at Sea AI Department is dedicated to pushing the boundaries of visual synthesis. We aim to achieve industry leadership in high-fidelity portrait and video generation. This team focuses on fundamental research and the scaling of generative models to empower next-generation social and E-commerce platforms.
About The Job
- Architecture Design: Lead the architecture design and implementation of video generation post-training, focusing on high-quality instruction data, preference alignment, and video quality enhancement.
- Capability Expansion: Explore long-video modeling, storyline consistency, and precise camera control.
- Alignment & Evaluation: Build video quality evaluation pipelines and multi-dimensional Reward Models; execute alignment training using RLHF, DPO, GRPO, and PPO.
- Inference Acceleration: Research and implement model distillation and other techniques for Diffusion model inference acceleration.
Requirements
- Education: Master’s or PhD in Computer Science or related fields. Bachelor's can be considered with a strong industrial experience.
- Technical Depth: Solid theoretical and practical foundation in video generation post-training, including preference alignment (RLHF/DPO/GRPO/PPO), fine-tuning (LoRA/QLoRA/DoRA), and distillation (Consistency Models, Flow Matching).
- Project Experience: Proven track record in video generation post-training, with practical experience in Video Reward Models and quality evaluation metrics.
- Soft Skills: Excellent communication and teamwork abilities.
- Plus points
- End to end training: Experience in the full lifecycle of high-quality video generation model training.
- Frontier Research: Research leadership in physical simulation, world consistency, temporal consistency, or causal reasoning.
- Preference Alignment: Expertise in high-quality video evaluation and human preference alignment.
- Efficiency Mastery: Experience in extreme efficiency optimization (inference acceleration, VRAM compression, distillation, quantization).