大模型推理研发工程师（J106548）
Core
End-to-end deployment and optimization of large language models (LLMs) and specialized models from weights to online services, focusing on inference acceleration, cost reduction, and stability for high-traffic products.
Role type
Senior IC LLM inference engineer (systems & optimization)
Builds
Production-grade inference services for text, image, and voice LLMs across multi-hardware platforms
Domain
AI / Large Language Models / High-performance Computing
Deliverable
production ML models
Required skills
C/C++, Python, CUDA programming, model quantization, KV Cache optimization, speculative decoding, continuous batching, custom operator development, heterogeneous hardware tuning (NVIDIA, Kunlun, Ascend), distributed inference (tensor/pipeline parallelism), multi-modal model inference
Preferred skills
Experience with trillion-parameter models, high-concurrency online services, SGLang, TensorRT-LLM, domestic chip adaptation (Kunlun, Ascend)
Technologies
PyTorch, SGLang, TensorRT-LLM, NVIDIA GPUs, Kunlun, Ascend
Responsibilities
Optimize inference engines for core scenarios to reduce TTFT and increase throughput; develop custom operators and tune performance on heterogeneous hardware; design and validate multi-modal inference solutions; manage memory and scheduling strategies for stable online services
Seniority
Senior, hands-on IC
