Senior AI System Engineer - LLM Training System
Core
Develop core compute operators and distributed scaling systems for foundation model pre-training and post-training on Ascend NPUs.
Role type
Senior IC AI system engineer (LLM training)
Builds
High-performance distributed training pipelines and operators for foundation models
Domain
AI infrastructure / Large Language Model training / Ascend NPU hardware
Deliverable
production ML models
Required skills
AscendC, C++, Python, Triton, TileLang, distributed training frameworks (Megatron-LM, PyTorch FSDP, vLLM), performance profiling, low-precision arithmetic (FP8, MXFP8)
Preferred skills
Ascend NPU microarchitecture knowledge, Triton/TileLang backend optimization, open-source contributions to AI acceleration projects
Technologies
AscendC, Triton, TileLang, MindSpeed, Megatron-LM, PyTorch, vLLM, Ascend Profiler
Responsibilities
Design and tune core operators (FlashAttention, GEMM, MoE routing) using AscendC and Triton; Architect and optimize distributed pre-training and RL post-training pipelines; Tune multi-dimensional hybrid parallelism (TP, PP, CP, EP) on NPU clusters; Implement low-precision training schemes ensuring numerical stability; Conduct deep performance profiling to eliminate hardware stalls; Contribute high-performance operators to open-source communities
Seniority
Senior, hands-on IC