多模态大模型训练加速工程师-Data AML(北京/上海/杭州/深圳)
Core
Accelerate post-training tasks (SFT, RL, CT, LoRA) for video/image generation and multimodal models, including heterogeneous GPU adaptation for Seedance/Seedream models.
Role type
Senior IC machine-learning engineer (training acceleration)
Builds
Optimized training pipelines for AIGC and multimodal models
Domain
AI/ML, Computer Architecture, Distributed Systems
Deliverable
production ML models
Required skills
AI engineering system optimization, computer architecture, visual large model principles, distributed training methods, AIGC algorithm principles, PyTorch, Megatron, DeepSpeed, ZeRO, TP, SP, Gradient Checkpoint, Activation Offloading, compute fusion, low-precision training, Triton, TileLang, Cute DSL, Verl, OpenRLHF, TRL
Preferred skills
Heterogeneous GPU adaptation experience, DSL optimization experience
Technologies
PyTorch, Megatron, DeepSpeed, Triton, TileLang, Cute DSL, Seedance, Seedream