大模型性能优化研发专家
Core
Develop and optimize training and inference frameworks for large models to meet extreme performance requirements for public cloud customers.
Role type
Senior IC large model performance optimization engineer
Builds
Accelerated training and inference frameworks for public cloud and internal core business
Domain
Cloud computing, large language models, high-performance computing
Deliverable
production ML models
Required skills
vLLM, SGlang, TensorRT-LLM, Megatron-LM, DeepSpeed, verL, Llama-factory, CUDA programming, NCCL, MPI, distributed parallelism strategies, operator fusion, communication optimization
Preferred skills
Autonomous driving algorithms (BEVformer, MapTRV2, SparseDrive, FlashOCC, Pointpillars), large project architecture design
Technologies
vLLM, SGlang, TensorRT-LLM, Megatron-LM, DeepSpeed, verL, Llama-factory, CUDA, NCCL, MPI
Responsibilities
Develop acceleration schemes based on cloud platforms and industry frameworks; Analyze performance bottlenecks in key POCs and resolve critical issues; Track academic trends to integrate latest research into framework features; Support delivery and iteration of self-developed core business training/inference frameworks.
