All rights reserved. Amazon Trademark Deploy GluonCV Models GluonCV Models MXNet Computational Graph Json Acyclic Graph Export As-is Optimize with TVM© 2019, Amazon Web Services, Inc. or its Affiliates its Affiliates. All rights reserved. Amazon Trademark Like GluonCV? Go build! https://gluon-cv.mxnet.io https://github.com/dmlc/gluon-cv© 2019, Amazon Web Services, Inc. or its Affiliates. All rights

ingests a FP32 graph and a small dataset • Finds suitable quantization scale • Produces a quantized graph • Compiling Pre-quantized models – QNN Dialect • TVM ingests a pre-quantized graph in TFLite or or MxNet • Use high-level wrapper ops of QNN dialect© 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. TVM Overview Framework Graph Mxnet TF …. parsers Relay Graph Target-independent Target-independent Relay passes Target-optimized graph Target-dependent Relay passes Intel x86 ARM CPU Nvidia GPU ARM GPU Schedule templates written in TVM Tensor IR .. More targets AutoTVM – Tuning

can run any models Your compiler (TVM) supports multiple frontends (e.g., TensorFlow, PyTorch, MXNet) Non Maximum Suppression ResNet-50 Dense Your Chip Your Chip© 2019, Amazon Web Services, Inc System Overview Relay IR Graph Annotation with Your Annotator Graph Partitioning Your Codegen LLVM, CUDA, Metal, VTA Serialized Subgraph Library Relay Runtime (VM, Graph Runtime, Interpreter) Mark supported operators or subgraphs 1. Implement an operator-level annotator, OR 2. Implement a graph-level annotator© 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Option 1:

Inference Flow >> 5 MxNet CPU Layers FPGA Layers Runtime Image Model Weights Calibration Set Quantizer Compiler Tensor Graph Optimization Framework Tensor Graph to Xilinx Tensor Graph Frontend Deep https://github.com/xilinx© Copyright 2018 Xilinx TVM as Unified ML Front End >> 6 Relay (and NNVM) Graph Parser XIR Compiler Quantizer Partitioner @relay.transform.module_pass(opt_level=4) class AccelModule:© supported/not supported, pattern matching graph colorization - Choices how to partition especially for multi-branch networks (i.e. YOLOv3, SSD)© Copyright 2018 Xilinx TVM Graph Partitioning/Fusion >> 8 Subgraph

dependent: arange, nms, etc. ○ Control flow: concatenate within a while loop Limitation of TVM/graph runtime ● Cannot compile and run dynamic models© 2019, Amazon Web Services, Inc. or its Affiliates new runtime for Relay ● Dynamic codegen (WIP) ○ Kernel dispatch for a single op ○ Graph dispatch for a (sub-)graph In collaboration with Jared Roesch, Zhi Chen, Wei Chen© 2019, Amazon Web Services, implement© 2019, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Dispatch a Whole Graph Resnet Data -> (Any, 3, 224, 224) Dispatch Tree Resnet_copy0 Resnet_copy1 ... 1 <= bs < 17 17

Hexagon DSP 人NiOS ! 驱动万物知 Tensorflow deploy.so / deploy.json / deploy.bin | NNVM / Relay 让 Graph Optimization 站 站 Compile | libtvm_hexagon_runtime.so Alios TVM @ Hexagon DSP 。 Compute Kernel 6 4 2.353 2. ， 曾硬证 0 Mobilenet 1.0 densenet121 量TVM (with Auto Tuning) 目MXNet+ TensorRT 目TVM +TensorRT AiiOS ! 驱动万物智能 THANKS9 Ali0S ! 驱动万物智能

and partly because of a strong focus on performance from the inception of the project, Julia's computational efficiency exceeds that of other dynamic languages, and even rivals that of statically-compiled that are natively supported on modern computers, thus allowing Julia to take full advantage of computational resources. Additionally, Julia provides software support for Arbitrary Precision Arithmetic, all values in Julia are true objects having a type that belongs to a single, fully connected type graph, all nodes of which are equally first-class as types. 120CHAPTER 11. TYPES 121 • There is no meaningful

and partly because of a strong focus on performance from the inception of the project, Julia's computational efficiency exceeds that of other dynamic languages, and even rivals that of statically-compiled that are natively supported on modern computers, thus allowing Julia to take full advantage of computational resources. Additionally, Julia provides software support for Arbitrary Precision Arithmetic, all values in Julia are true objects having a type that belongs to a single, fully connected type graph, all nodes of which are equally first-class as types. 120CHAPTER 11. TYPES 121 • There is no meaningful

arithmetic calculation involving decimal numbers. In AI, total FLOPs are often used to estimate the computational cost of training or running a model. Note: Only language models shown (per Epoch AI, includes development – one that builds on recent exponential gains in model scale, training data, and computational efficiency. Timelines for AGI remain uncertain, but expert expectations have shifted forward scale and sophistication of artificial intelligence is demanding an extraordinary amount of computational horsepower, primarily from AI-focused data centers. These facilities – purpose-built to train

and partly because of a strong focus on performance from the inception of the project, Julia's computational efficiency exceedsCHAPTER 1. JULIA 1.11 DOCUMENTATION 4 that of other dynamic languages, and that are natively supported on modern computers, thus allowing Julia to take full advantage of computational resources. Additionally, Julia provides software support for Arbitrary Precision Arithmetic, all values in Julia are true objects having a type that belongs to a single, fully connected type graph, all nodes of which are equally first-class as types. 127CHAPTER 12. TYPES 128 • There is no meaningful