Research Engineer (Triton)

Proficient in PyTorch or mainstream inference frameworks (e.g., Megatron-LM, TensorRT). This role focuses on the research and productionisation of automated inference algorithms — designing core algorithms for automated deployment, automated compilation optimization, and adaptive scheduling — using algorithmic approaches to streamline the inference delivery pipeline, maximize single-GPU throughput, and improve multi-GPU and cluster-level efficiency.ResponsibilitiesStudy the underlying architecture of inference frameworks and design end-to-end automation solutions for the inference pipeline.Develop algorithms for automated service configuration, distributed inference sharding, and dynamic scheduling.Research automatic kernel generation and hardware-adaptive optimization based on the Triton Compiler.Implement auto-tuning algorithms such as automatic quantization and operator fusion, and build a closed-loop performance benchmarking system.Track cutting-edge research in AI compilation and automated inference, and deploy in-house algorithms to the production inference platform.RequirementsMaster's degree or above in Computer Science, AI, or High-Performance Computing; solid command of the deep learning inference pipeline.Proficient in C++/Python, Linux, and CUDA development; familiar with mainstream inference architectures.Understanding of the Triton Compiler pipeline; experience in AI compilation or operator optimization is a plus.Bonus: experience with PyTorch graph-mode tuning, pytorch-inductor, automated model compression, TVM/MLIR, or large-model inference in production.