异构计算与Token效能架构师 - TikTok研发
Core
Design and evolve end-to-end architecture for AI heterogeneous computing infrastructure, optimizing the efficiency from Token demand to GPU resource delivery for TikTok's core business scenarios.
Role type
Senior IC Heterogeneous Computing & Token Efficiency Architect
Builds
Unified resource models, scheduling systems, and observability platforms for LLM inference and GPU clusters
Domain
AI Infrastructure / Large Language Models / Heterogeneous Computing
Deliverable
production ML models
Required skills
System design, distributed systems, LLM inference concepts, resource scheduling, GPU architecture, data analysis
Preferred skills
LLM Serving, Token efficiency optimization, GPU virtualization, CUDA/NCCL, AI cost optimization
Technologies
Go, C++, Python, Kubernetes, Slurm, CUDA, NCCL, RDMA, PyTorch, vLLM, SGLang, TensorRT-LLM
Responsibilities
Design architecture connecting applications, agents, and models to GPU resources; Build Token efficiency metrics and optimization systems; Optimize model selection and routing for cost/performance; Unify and pool heterogeneous GPU resources; Establish capacity governance for Token and compute; Create unified observability for Token and infrastructure efficiency.
Seniority
Senior, hands-on IC