人脸分类模型: 超大softmax  3D卷积模型 M6模型 RapidFormer性能 工程优化: 模型并行(Whale)  FP16 / Int8  模型剪枝  Op融合(Fusion Stitch)  MILR: Blade Disc 工程优化: Blade模型推理 Dynamic Shape Compiler for Machine Learning Workloads Graph去重] 内存Allocate优化 ParallelStringOp [split/type conversion] Sequence Feature [side info] Op Fusion [hash + embedding] Overlap Execution [FG OP化] Item Feature增量更新 3.工程优化复 杂 4.数据获取困 难 挑战 深度模型是非线性的: Video Fram es Bert Title OCR Cls Tok en Title feature OCR feature Im age feature M HSA Fusion M VM VTM M TM Tran sform er decoder Tran sform er decoder Tran sform er decoder Tran

������ ���� ���� 主题分类-特征提取 DPN SENet ResNeXt NASNet 主题分类-模型训练 模型融合 a) Early fusion b) Late fusion 三段式方式 Loss Func� Loss ����� �� ü ���� ü ����� ü ���� ü ���� � ü ��� ü ���� 人物信息结构化

算法模型层 4 深度学习-分布式模型推理 • 推理性能优化 • 减少计算量： operator fusion/XLA/TVM/prune/float16/quantization • 加快计算速度： batching/TensorRT/MPS/SSE/AVX/Neon • operator fusion • 针对特定场景重写耗时算子 • 重构tensorflow计算引擎 • batching

training Semi-supervised Labeling Incremental training Data Augment Model comparison Model Fusion Gray Update Auto Evaluation Log Server Graph Abstraction Data Flow API Manager Pipeline

????????? ???????????? ????????????0 ???????????? ME ????????????????????????→0 SPMC Detail Fusion Net Our Method 48 ???????????????????????? ???????????? ????????????0 ???????????? ME ????

jit.script and torch.jit.tracepreview features including better support for pointwise operations in fusion. ‣ Added support for a C++ only API (new PyTorch 1.0 preview feature). ‣ Dataloader may still throw Reliability: Some cases where a dataloader could hang if shutdown during its iteration has been fixed. ‣ Fusion: Tensor and constant scalar operations, like add(t, 1), and chunk operations are now fusable. ‣