Chimera SDK Quick Start Guide

Welcome to the Chimera SDK Quick Start Guide! This guide will instruct you on how to download and install the Chimera SDK onto your local machine.

Quadric Developers Studio

NOTE: If this is your first experience with Quadric and the Chimera Processor family, we recommend starting your evaluation with the Quadric Developer Studio.

DevStudio has an easy-to-use graphical interface where you can mix-and-match optimized libraries of C++ pre-processing and post-processing kernels with a variety of pre-compiled Deep Neural Network (DNN) models. It's a great place to quickly prototype an application signal chain and familiarize yourself with the performance benefits of the Chimera processor architecture.

Prerequisites

Hardware Requirements

Quadric recommends a laptop, workstation, or virtual machine with the following specs:

Software Requirements

Quadric CLI Documentation The Quadric CLI tool provides a command line interface for interacting with the Quadric API and managing SDK operations. System Requirements The Quadric CLI is supported on Linux environments and requires either Docker or Podman installed as a container runtime. All SDK operations run within containers, ensuring a consistent development environment.

The Quadric CLI tool provides a command line interface for interacting with the Quadric API and managing SDK operations.

Quadric SDK cli Usage

quadric [command] [flags]

Global Flags

Available Commands

auth

Authenticate with the Quadric API. The auth command manages your authentication with the DevStudio account.

quadric auth [command] [flags]

Available Auth Commands

license

Manage machine license for the SDK.

quadric license [command] [flags]

Available License Commands

sdk

Execute SDK operations. The SDK command provides a comprehensive set of tools for managing and using the Quadric SDK.

quadric sdk [command] [flags]

Available SDK Commands

Interactive Sessions

The quadric sdk interactive command launches an interactive session inside the SDK container environment. This provides direct access to the SDK tools and development environment.

quadric sdk interactive

When you start an interactive session, you'll be placed in a container with the SDK tools and environment pre-configured. This environment includes:

• Access to SDK development tools
• Pre-configured build environment
• Tools for model compilation and testing
• Direct access to the SDK utilities

To exit the interactive session, simply type exit or use Ctrl+D.

Jupyter Notebook Support

The quadric sdk notebook command launches a Jupyter Lab server within the SDK environment, providing an interactive notebook interface for development and experimentation.

quadric sdk notebook

By default, the Jupyter server runs on port 8888. Depending on your environment, you may need to configure port forwarding to access the notebook interface.

When you start the notebook server, you'll see output similar to:

Using Docker version 24.0.6
Using SDK version 24.09

[I 2024-12-10 00:06:50.978 ServerApp] Jupyter Server 2.14.2 is running at:
[I 2024-12-10 00:06:50.978 ServerApp] http://hostname:8888/lab?token=<your_token>
[I 2024-12-10 00:06:50.978 ServerApp] http://127.0.0.1:8888/lab?token=<your_token>

To access the notebook:

1. Copy the URL provided in the output
2. Paste it into your web browser
3. The URL includes an authentication token that will log you in automatically

To exit the notebook server, use Ctrl+C in the terminal where it's running.

Graph Compilation

The sdk graph compile command provides options for compiling ONNX models.

quadric sdk graph compile [OPTIONS] QUANTIZED_MODEL

Graph Compile Options

Example Graph Compilation

## Basic compilation
quadric sdk graph compile model.onnx

## Compilation with specific target and memory
quadric sdk graph compile --target QB16 --ocm-size 8MB model.onnx

## Compilation with performance validation
quadric sdk graph compile --run --validate-iss model.onnx

update

Check for and update to the latest version of the Quadric CLI tool.

quadric update

When updates are available, the CLI will notify you with a message like:

An update is available! Use `sudo quadric update` to install v0.0.22

Container Runtime Selection

The SDK commands that use containers (interactive, notebook, etc.) can be forced to use either Docker or Podman:

## Force using Docker
quadric sdk [command] --force-docker

## Force using Podman
quadric sdk [command] --force-podman

Getting Help

## Get general help
quadric --help

## Get command-specific help
quadric <command> --help

## Examples:
quadric auth --help
quadric license --help
quadric sdk --help
quadric sdk graph compile --help

Install the Chimera Host APIs

Login to your Quadric DevStudio account at https://app.quadric.ai/login.

Click on the Resources tab in the taskbar. In the left-hand menu, select SDKs.

Click on the Download drop-down button next the Quadric Chimera SDK version you wish to download, then select the Bindings.

Read and accept the Software User License Agreement. A file named bindings-YYYY-MM-DD.zip will begin downloading.

NOTE: you will need to change the YYYY-MM-DD text in the commands below to match the version of the Chimera SDK that you've downloaded. In the image above, the SDK version is 2023.01.1.

Unzip the downloaded file. Inside will be two more zip files, one for each of the supported host platforms, Linux and macOS:

$ unzip bindings-YYYY.MM.DD.zip
Archive:  bindings-YYYY.MM.DD.zip
  inflating: quadric_python_bindings_V2_Linux_X64.zip
  inflating: quadric_python_bindings_V2_macOS_X64.zip

Unzip the appropriate zip file for your host platform and install the Python wheel (*.whl) file matching the version of Python in your local environment:

$ unzip quadric_python_bindings_V2_macOS_X64.zip
Archive:  quadric_python_bindings_V2_macOS_X64.zip
  inflating: quadric-0.0.1-cp310-cp310-macosx_12_0_x86_64.whl
  inflating: quadric-0.0.1-cp38-cp38-macosx_12_0_x86_64.whl
  inflating: quadric-0.0.1-cp39-cp39-macosx_12_0_x86_64.whl

$ pip install quadric-0.0.1-cp39-cp39-macosx_12_0_x86_64.whl

Next Steps

Now that you have the Chimera SDK installed on your local machine, check out the available commands in the Chimera SDK Command Line Interface (CLI), the easiest way to interact with the components that make up the Chimera SDK.

Once you're familiar with the Chimera SDK CLI, try running one of one of our tutorials:

• Tutorial: Creating a Simple Kernel

• Tutorial: Writing a Kernel in C++ API

• Tutorial: Custom Operator Insertion into a Model

• Tutorial: Custom Operator Insertion for Operators without an ONNX Equivalent