Deploying Large Language Models on Beijing Xunwei iTOP-RK3588

Deploying Large Language Models on Beijing Xunwei iTOP-RK3588

The field of artificial intelligence (AI) has seen a growing interest in large models.Large models, as the name suggests, refer to deep learning models with a vast number of parameters and highly complex structures.The emergence of these models has not only propelled breakthroughs in AI technology but has also brought revolutionary changes to various industries.

The RK3588 is a next-generation flagship high-end processor launched by Rockchip, designed with a 8nm process, featuring an octa-core CPU with four A76 and four A55 cores, and an Arm high-performance GPU, along with an NPU that has 6T computing power. This provides robust hardware support for the operation of large models.

Deploying Large Language Models on Beijing Xunwei iTOP-RK3588

On the iTOP-RK3588 development board launched by Beijing Xunwei, the RKLLM toolkit supports the conversion and deployment of LLM (large language model) on the iTOP-RK3588 platform. It is compatible with the Hugging Face model architecture (Hugging Face is widely used for natural language processing tasks). Currently, the system supports the following models: LLaMA, Qwen, Qwen2, and Phi-2. It supports quantization techniques, specifically using w8a8 (8-bit weights, 8-bit activations) and w4a16 (4-bit weights, 16-bit activations) precision for model quantization. This can store and compute models more efficiently on the target platform, reducing memory usage.

Deploying Large Language Models on Beijing Xunwei iTOP-RK3588

Experience with the iTOP-RK3588 development board and large models

To allow users to experience RKLLM more quickly, Beijing Xunwei has upgraded the NPU version in the Linux kernel source code to the latest version 0.9.6, as detailed below:

Deploying Large Language Models on Beijing Xunwei iTOP-RK3588

By default, the RKLLM dynamic library is integrated into Ubuntu and Debian systems, allowing users to directly copy the converted RKLLM large prediction models and inference programs for testing. For detailed instructions, please refer to the NPU manual for the relevant steps on RKLLM model conversion and testing.

Update Content

iTOP-RK3588 Development BoardNPU User Manual》v1.1

Added Chapter 8 related to RKLLM large language model testing

How to obtain: Contact customer service to join the RK3588 after-sales group

Purchase link:

https://m.tb.cn/h.5EZ0ev1LMbQdsiI?tk=rrl8WmlLtmr

Tutorial Directory

Chapter 1 Hello! NPU

1.1 The Birth of NPU!

1.2 Getting to Know RKNPU

Chapter 2 Preparing the RKNPU Development Environment

2.1 Development Environment

2.1 Software Architecture

2.2 SDK Description

Chapter 3 Getting NPU Running

3.1 Using NPU in Linux

3.1.1 Setting Up Cross Compiler

3.1.2 Modifying Compiler Tool Path

3.1.3 Updating RKNN Model

3.1.4 Compiling Demo

3.1.5 Running Demo on Development Board

3.2 Using NPU in Android

3.2.1 Downloading Required Tools for Compilation

3.2.2 Modifying Compiler Tool Path

3.2.3 Updating RKNN Model

3.2.4 Compiling Demo

3.2.5 Running Demo on Development Board

Chapter 4 Experience RKNN_DEMO

4.1 rknn_ssd_demo Experience

4.2 rkn_api_demo Experience

4.3 rknn_multiple_input_demo Experience

Chapter 5 Model Conversion

5.1 RKNN-Toolkit2 Introduction

5.2 Setting Up RKNN-Toolkit2 Environment

5.2.1 Installing Miniconda

5.2.2 Creating RKNN Virtual Environment

5.2.3 Installing Pycharm

5.2.4 Configuring Pycharm

5.3 Using RKNN-Toolkit2 Tools

5.3.1 Running Model in Emulator

5.3.2 Running Model on RK3588 Development Board

Chapter 6 Other Model Conversions

6.1 Using TensorFlow Framework

6.2 Using Caffe Framework

6.3 Using TFLite Framework

6.4 Using ONNX Framework

6.5 Using Darknet Framework

6.6 Using PyTorch Framework

Chapter 7 Using RKNN-Toolkit-lite2

7.1 Main Features Description

7.2 Steps to Set Up Environment

7.2.1 Installing Miniconda

7.2.2 Creating RKNN Virtual Environment

7.2.3 Installing RKNN-ToolkitLite 2 Package

7.2.4 Installing OpenCV

7.3 Running Test Program

Chapter 8 RKLLM Large Prediction Model Testing

8.1 Introduction to RKLLM-Toolkit

8.2 Setting Up RKLLM-Toolkit Environment

8.2.1 Installing Miniconda

8.2.2 Creating RKLLM Virtual Environment

8.3 Large Language Model Conversion

8.4 Compiling Inference Program

8.5 Running Tests on Development Board

Video Showcase

■ Follow the “Xunwei Electronics” WeChat public account for tutorials, materials, industry insights, and product resources.

■ For more information about Xunwei, please contact us:

Sales Engineer: 010-8527-0708 Extension 1

Technical Support: 010-8527-0708 Extension 2

Custom Solutions: 010-8527-0708 Extension 3

■ Real-time technical support:

AM 9:00—12:00, PM 13:30—17:30 (Monday to Saturday)

Technical Group Chat【824412014

END

Long press to recognize the QR code to follow

Deploying Large Language Models on Beijing Xunwei iTOP-RK3588

Xunwei Electronics

Making learning easier and development simpler

Deploying Large Language Models on Beijing Xunwei iTOP-RK3588
http://www.topeet.com/

Deploying Large Language Models on Beijing Xunwei iTOP-RK3588

Deploying Large Language Models on Beijing Xunwei iTOP-RK3588 Deploying Large Language Models on Beijing Xunwei iTOP-RK3588

【RK3588 Adaptation to Android 13】 【Hongmeng Learning Development Tutorial】

Deploying Large Language Models on Beijing Xunwei iTOP-RK3588 Deploying Large Language Models on Beijing Xunwei iTOP-RK3588

【RKNPU2 Three Major Projects in Practice】 【RK3568 Driver Guide Issue 14】

Leave a Comment

Your email address will not be published. Required fields are marked *