Software Installation

Software Installation#

Acquiring Software#

Visit the BIDL open-source page to obtain the source code project.

Installation Package

Clone the LynxiTech/BIDL repository to local directory, such as:

Note that this package only include project source code; for code folder details, refer to Directory Description.

Resources Package

The resource package includes compressed files, which can be downloaded from Baidu Net-disk, after download, please unzip them to the same directory as in the baidu net-disk.

The resource package includes the dataset, pre-trained weights (executed in GPU/CPU) and compiled model files (executed in lynxi chip) for BIDL, including:

  • data: datasets;

  • model_files: model files;

  • weight_files: weight files.

Deploy Training Environment#

The GPU training environment this framework depends on is mainly PyTorch, so PyTorch needs to be installed first.

First, install the GPU version of PyTorch.

pip install torch torchvision torchaudio

Deploy Compilation Environment#

Since Lyngor, which is used for compiling models, only supports the CPU version of PyTorch, you need to reconfigure the CPU version of the PyTorch environment. Additionally, compilation and deployment require support from Lyngor and SDK software, as well as APU hardware.

Execute the following command to install the latest stable version of PyTorch provided by the PyTorch official website.

pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cpu

Validation#

GPU Environment Validation#

Run the pip list command to check whether the installed packages fully include those specified in requirements.txt.

Import the corresponding packages in the Python environment to verify correct installation.

>>> torch.__version__
'2.4.0+cu121'

CPU Environment Validation#

Run the pip list command to check whether the installed packages fully include those specified in requirements.txt.

Import the corresponding packages in the Python environment to verify correct installation.

>>> torch.__version__
'2.4.0+cpu'

Directory Description#

After BIDL is installed, refer to the table below for the directory structure and description.

Table: BIDL Directory Structure

Directory/File

Description

applications

The storage directory for application cases, which includes the following categories:

  • classification (Classification)

  • dvsdetection (Detection)

  • neuralsim (Brain-like Neuron and SNN Simulation)

  • onlinelearning (Online Learning)

  • videotracking (Visual Tracking)

  • robotics (Robotics)

  • functionalBII (Functional Level Brain-Inspired Intelligence)

Among them, functionalBII and onlinelearning are only provided in the professional version.

lynadapter

Directory for storing Lynxi chip interface files

requirements

Specifies third-party software packages required for the project

tools

Includes training and inference scripts

tutorial

Path for storing demonstration cases

utils

Path for storing utility functions

deploy

Path for storing demos that operate without the framework
Contents