AI Algorithm Application Engineer AI演算法應用工程師

1. AI Operator Adaptation Development: Implement custom operator development for frameworks like NVIDIA and Horizon, or perform equivalent refactoring for incompatible operators;
2. Develop Heterogeneous Computing Pipelines: Achieve coordinated inference across devices including CPUs, GPUs, DLAs, and NPUs;
3. Performance Bottleneck Analysis: Establish a three-dimensional evaluation model for compute power, latency, and power consumption to identify computational bottlenecks;
4. Model lightweighting: Optimize deep learning models using techniques like quantization, pruning, sparsification, and structural replacement to reduce computational demands and power consumption. Develop generic frameworks for rapid adaptation to diverse task models;
5. Pre/post-processing optimization: Enhance code efficiency through CUDA/OpenCL implementation;

1. AI算子適配開發：實現NVIDIA、地平線等框架的Custom OP開發或實現不相容算子的等效重構；
2. 開發異構計算流水線：實現CPU、GPU、DLA、NPU等設備的協調推理；
3. 性能瓶頸分析：建立算力-時延-功耗三維評估模型，識別計算瓶頸；
4. 模型輕量化改造：利用量化、剪枝、稀疏、結構替換等模型優化方法，對深度學習模型進行輕量化處理，降低算力需求與功耗。整理通用框架，實現對各任務模型的快速適配；
5. 前後處理優化：利用CUDA/OpenCL實現對代碼效率的提升；