important things: • torch.no_grad() • Don’t store the history of all computations • eval() • Tell compiler which mode to run on. Visualization • TensorboardX (visualise training) • PyTorchViz (visualise

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 EmbeddingVariable [No Hash Conflict] 特征准入/淘汰 Adaptive Embedding

out of 4 contiguous values in a matrix are 0 (effectively 50% sparsity). The intermediate model compiler rewrites a standard matrix multiplication operation to be performed using a compressed representation