Job opportunity
Senior Ascend NPU Systems Engineer
Huawei Canada
Ascend NPU systems engineering drives Huawei Canada’s foundation-model performance through AscendC kernels and distributed scaling. This is a permanent senior engineering opportunity.
As a Senior System Engineer, you will develop compute operators and optimize pre-training and post-training workloads. The work spans AscendC, Triton, TileLang, MindSpeed, and upstream frameworks. You will improve training efficiency across Ascend NPU clusters through profiling, parallelism, and numerical alignment.
Key Responsibilities:
Design and tune FlashAttention, GEMM, RMSNorm, and MoE operators
Optimize MindSpeed and distributed foundation-model training pipelines
Tune TP, PP, CP, EP, ZeRO, and FSDP parallelism
Profile Ascend hardware and reduce compute-communication overhead
Contribute operators and acceleration features to open-source projects
Requirements:
Technical degree and three-plus years of relevant experience
AscendC or CUDA C/C++ operator-development experience
Distributed framework experience with MindSpeed, Megatron, FSDP, veRL, or vLLM
Strong C/C++ and modern Python programming proficiency
Experience with Ascend Profiler, PyTorch Profiler, or Nsight
Apply systems expertise in AscendC, distributed training, profiling, and low-precision computing to Huawei Canada’s accelerator platforms.
#J-18808-Ljbffr
Apply on source site As a Senior System Engineer, you will develop compute operators and optimize pre-training and post-training workloads. The work spans AscendC, Triton, TileLang, MindSpeed, and upstream frameworks. You will improve training efficiency across Ascend NPU clusters through profiling, parallelism, and numerical alignment.
Key Responsibilities:
Design and tune FlashAttention, GEMM, RMSNorm, and MoE operators
Optimize MindSpeed and distributed foundation-model training pipelines
Tune TP, PP, CP, EP, ZeRO, and FSDP parallelism
Profile Ascend hardware and reduce compute-communication overhead
Contribute operators and acceleration features to open-source projects
Requirements:
Technical degree and three-plus years of relevant experience
AscendC or CUDA C/C++ operator-development experience
Distributed framework experience with MindSpeed, Megatron, FSDP, veRL, or vLLM
Strong C/C++ and modern Python programming proficiency
Experience with Ascend Profiler, PyTorch Profiler, or Nsight
Apply systems expertise in AscendC, distributed training, profiling, and low-precision computing to Huawei Canada’s accelerator platforms.
#J-18808-Ljbffr