TensorFlow 1.x Version (vai_q_tensorflow) - 1.4.1 English - UG1414

Vitis AI User Guide (UG1414)

Document ID: UG1414
Release Date: 2021-12-13
Version: 1.4.1 English

Revision History
Vitis AI Overview
Navigating Content by Design Process
Features
Vitis AI Tools Overview
Deep-Learning Processor Unit
DPU Naming
Zynq UltraScale+ MPSoC: DPUCZDX8G
Alveo U50/U280 Card: DPUCAHX8H
Alveo U50/U50LV/U280 Card: DPUCAHX8L
Alveo U200/U250 Card: DPUCADF8H
Versal AI Core Series: DPUCVDX8G
Versal AI Core Series: DPUCVDX8H
Vitis AI Model Zoo
Vitis AI Optimizer
Vitis AI Quantizer
Vitis AI Compiler
Vitis AI Profiler
Vitis AI Library
Vitis AI Runtime
Vitis AI Containers
Minimum System Requirements
Development Flow Overview
Getting Started
Installation and Setup
Downloading Vitis AI Development Kit
Setting Up the Host
Installing the Tools
Setting Up the Host (Using VART)
For Edge
For Cloud
Setting Up the Evaluation Board
Setting Up the ZCU102/ZCU104/KV260/VCK190 Evaluation Board
Flashing the OS Image to the SD Card
Booting the Evaluation Board
Accessing the Evaluation Board
UART Port
Using the Ethernet Interface
Using the Board as a Standalone Embedded System
Installing Vitis AI Runtime on the Evaluation Board
Setting Up the Custom Board
Running Examples
Vitis AI Examples
Running Vitis AI Examples
For Edge (DPUCZDX8G/DPUCVDX8G)
For Cloud
Support
Quantizing the Model
Overview
Vitis AI Quantizer Flow
TensorFlow 1.x Version (vai_q_tensorflow)
Installing vai_q_tensorflow
Running vai_q_tensorflow
Preparing the Float Model and Related Input Files
Generating the Frozen Inference Graph
Preparing the Calibration Dataset and Input Function
Quantizing the Model Using vai_q_tensorflow
Generating the Quantized Model
(Optional) Evaluating the Quantized Model
(Optional) Dumping the Simulation Results
vai_q_tensorflow Quantization Aware Training
Generated Files
QAT APIs for TensorFlow 1.x
tf.contrib.decent_q.CreateQuantizeTrainingGraph(config)
tf.contrib.decent_q.CreateQuantizeEvaluationGraph(config)
tf.contrib.decent_q.CreateQuantizeDeployGraph(checkpoint, config)
Tips for QAT
vai_q_tensorflow Supported Operations and APIs
vai_q_tensorflow Usage
TensorFlow 2.x Version (vai_q_tensorflow2)
Installing vai_q_tensorflow2
Running vai_q_tensorflow2
Preparing the Float Model and Calibration Set
Quantizing Using the vai_q_tensorflow2 API
vai_q_tensorflow2 Fast Finetuning
Saving the Quantized Model
(Optional) Evaluating the Quantized Model
(Optional) Dumping the Simulation Results
vai_q_tensorflow2 Quantization Aware Training
Quantizing with Custom Layers
vai_q_tensorflow2 Supported Operations and APIs
vai_q_tensorflow2 Usage
vitis_quantize.VitisQuantizer
vitis_quantize.VitisQuantizer.quantize_model
vitis_quantize.VitisQuantizer.dump_model
vitis_quantize.VitisQuantizer.get_qat_model
vitis_quantize.VitisQuantizer.get_deploy_model
Examples
PyTorch Version (vai_q_pytorch)
Installing vai_q_pytorch
Running vai_q_pytorch
Preparing Files for vai_q_pytorch
Modifying the Model Definition
Adding vai_q_pytorch APIs to Float Scripts
Running Quantization and Getting the Result
Module Partial Quantization
vai_q_pytorch Fast Finetuning
vai_q_pytorch QAT
vai_q_pytorch QAT Requirements
vai_q_pytorch Usage
class torch_quantizer()
def export_quant_config(self)
def export_xmodel(self, output_dir, deploy_check)
Caffe Version (vai_q_caffe)
Installing vai_q_caffe
Running vai_q_caffe
vai_q_caffe QAT
vai_q_caffe Usage
Compiling the Model
Vitis AI Compiler
Compiling with an XIR-based Toolchain
XIR
Compiling for DPU
Compiling for Customized Accelerator
Supported Operators and DPU Limitations
Currently Supported Operators
Operators Supported by TensorFlow
Operators Supported by Caffe
Operators Supported by PyTorch
VAI_C Usage
Deploying and Running the Model
Programming with VART
C++ Example
Python Example
DPU Debug with VART
TensorFlow Workflow
Caffe Workflow
PyTorch Workflow
Multi-FPGA Programming
XRM
Multi-FPGA, Multi-Graph Deployment with Vitis AI
AI Kernel Scheduler
Apache TVM, Microsoft ONNX Runtime, and TensorFlow Lite
Apache TVM
Microsoft ONNX Runtime
TensorFlow Lite
Profiling the Model
Vitis AI Profiler
Vitis AI Profiler Architecture
Vitis AI Profiler GUI Overview
Getting Started with the Vitis AI Profiler
Installing the Vitis AI Profiler
Starting a Simple Trace with vaitrace
VAI Trace Usage
Configuration
Text Summary
DPU Profiling Examples
Optimizing the Model
Integrating the DPU into Custom Platforms
VART Programming APIs
C++ APIs
create_runner
create_runner
execute_async
wait
get_tensor_format
get_input_tensors
get_output_tensors
get_location
get_tensor
data
data_phy
sync_for_read
sync_for_write
copy_from_host
copy_to_host
copy_tensor_buffer
get_inputs
get_outputs
create_graph_runner
Python APIs
create_runner
execute_async
wait
get_input_tensors
get_output_tensors
get_inputs
get_outputs
create_graph_runner
Additional Resources and Legal Notices
Xilinx Resources
Documentation Navigator and Design Hubs
References
Please Read: Important Legal Notices

Back to home page