异构计算工程师 - 加速方向（J79682）
Core
Optimize training and inference efficiency for advanced models (LLMs, long sequences, multi-modal, MoE) on self-developed chips and large-scale heterogeneous clusters.
Role type
Senior IC heterogeneous computing engineer (acceleration)
Builds
Optimized training/inference pipelines for large models and autonomous driving models on custom silicon
Domain
AI/ML infrastructure + Semiconductor hardware
Deliverable
production ML models
Required skills
LLM/autonomous driving model architecture, PyTorch, Megatron, vLLM, GPU chip adaptation, heterogeneous cluster optimization, automatic parallelization, multi-chip mixed training
Preferred skills
Self-developed chip optimization experience
Responsibilities
Adapt common large models and autonomous driving models to self-developed chips; explore frontier technologies like automatic parallelization and multi-chip mixed training for large-scale heterogeneous clusters; collaborate with business teams to drive technical innovation.
