多模态大模型优化工程师-Data AML(北京/上海/杭州/深圳)
Core
Optimize self-developed and open-source multimodal large models (Seedance, Seedream, Seed-VLM) for hardware/software synergy, quantization, sparsity, and inference acceleration across distributed training and edge devices.
Role type
Senior IC multimodal large model optimization engineer
Builds
Training-free and post-training acceleration algorithms, efficient edge SLLM/SVLM models for Douyin/Jianying products
Domain
Generative AI, Large Language Models (LLM), Vision-Language Models (VLM), Distributed Systems
Deliverable
production ML models
Required skills
Python, PyTorch, CUDA/Triton, Quantization, Sparsity, MoE compression, Token compression, KV Cache optimization, Speculative decoding, Diffusion model optimization, Visual Transformer architecture, Distributed inference
Preferred skills
Papers in CVPR/ICCV/NeurIPS, Deep research in Diffusion models or Generative AI, Operator optimization, Multi-card parallelism
Responsibilities
Develop quantization and sparse acceleration algorithms for heterogeneous platforms; Optimize model representation dimensions (bit-width, resolution, steps) to reduce training/inference costs; Research efficient AIGC/LLM techniques using data synthesis and preference alignment; Engineer edge-side SLLM/SVLM solutions for global user products.
Seniority
Senior, hands-on IC