AI异构计算优化专家 - Data AML
Core
Evaluate and optimize heterogeneous computing chips for training and inference systems serving recommendation, advertising, CV, voice, and NLP business lines.
Role type
Senior IC heterogeneous computing optimization engineer
Builds
High-performance training and inference systems for internal business units and external enterprise clients via Volcano Engine
Domain
AI infrastructure, heterogeneous computing, machine learning systems
Deliverable
production ML models
Required skills
C/C++, Python, Linux, TensorFlow/PyTorch, deep learning models (GPT, SD, DiT, W&D), parallel computing architectures, high-performance operators, AI compilers (XLA, TVM, MLIR), GPU architecture and software stacks (CUDA, cuBLAS, CUTLASS)
Preferred skills
inference/training/communication optimization, SIMD/SIMT models, model pruning/quantization/LLM speculative sampling, Torch2.0+ compilation stack, Triton
Responsibilities
Evaluate heterogeneous computing chips and establish assessment frameworks; adapt chips for training and inference to reduce latency and increase throughput; develop high-performance operators; implement efficient heterogeneous hardware programming paradigms via compilation; research and validate emerging hardware-software directions like sparse computing and in-memory computing
Seniority
Senior, hands-on IC
