Multimodal Model Training and Inference Optimization Engineer

Work on ByteDance’s Vision-Applied Research team to improve training and inference efficiency for large-scale multimodal generative AI models. You’ll optimize large-model training pipelines, develop distributed training strategies (data/model/pipeline parallelism and communication), and benchmark/profile deep learning systems to remove performance bottlenecks. The role emphasizes scalable acceleration for models powering products like TikTok, CapCut, and Lemon8.