Llama3-8B-Instruct-Chinese Quick Deployment

Introduction

Llama3 is trained on a dataset of 150 trillion tokens by Meta, 7 times that of Llama2, including 4 times the code data. Among them, the pre-training dataset also includes 5% of non-English datasets, supporting up to 30 languages ​​in total, which will have more advantages in aligning other language capabilities. Llama 3 Instruct is even optimized for dialogue applications, combined with more than 10 million manually annotated data, and trained through Supervised Fine-Tuning (SFT), rejection sampling, Proximal Policy Optimization (PPO), and Direct Policy Optimization (DPO). The model mentioned in this article is the model that has been fine-tuned for Chinese instructions (Llama3-8B-Instruct-Chinese), which has relatively good performance in Chinese.

Quick Deployment

Log in to the UCloud Global console (https://console.ucloud-global.com/uhost/uhost/create), choose “GPU type” for the model, “High cost-effective graphics card 6”, and choose the detailed configuration such as the number of CPUs and GPUs as needed.
Minimum recommended configuration: 16-core CPU, 32G memory, 1 high cost-effective graphics card 6 GPU.
Select “Mirror Market” for the image, search for “Llama3” for the image name, and select this image to create a GPU cloud host.
After the GPU cloud host is created successfully, log in to the GPU cloud host.