MegaEase Cloud Major Updates GPU Services
Previously on
In April 2023, we made a major update to MegaEase Cloud, and in the last update, we made the following services available to the public：
- Infrastructure Software Updates
- Added Nginx and MinIO services
- Provided Kubernetes offline deployment
- Integrated with Cloudflare
- Zero Trust + MinIO CDN capability
- Easegress with CDN’s canary release and tracing capabilities
- Monitoring Metrics Collation and Optimization
- Benchmarking against DataDog and Prometheus
- Continuous integration capabilities
- Providing Docker image repository services
- Continuously application deployment through GitHub Actions
Today, MegaEase Cloud is getting a major feature update - GPU services.
We know that with today’s AIGC wave. Both programmers and ordinary people in general will want to embrace this wave. We want to touch AI, understand and experience AI, feel the impact of AI on our life, learning and entertainment, and even participate in it. We firmly believe that there will always be people in this world who are willing to keep exploring and accepting new things, and who are curious about the unknown areas of AI. However, the high cost of use (hardware and knowledge) is a barrier to exploration; however, we also believe that we have the ability to contribute to changing this situation through nearly 2 months of development and collaboration with our partners. We have completed the GPU service on MegaEase Cloud and polished many templates for AI lightweight training and inference (usage) scenarios for everyone to use. Now we are proud to announce that MegaEase Cloud’s GPU service is officially available for public beta. We hope you can enter the door of AI world through this service, feel the changes brought by AI and join the current AIGC wave, and become a participant instead of a spectator.
GPU Services
This service is currently only available on cloud.megaease.cn, cloud.megaease.com is not available at this time due to network reasons.
Select the GPU specification you need
On cloud.megaease.cn, we provide you with a platform to use GPUs at low cost through our GPU service. After logging into our platform, you can see the GPU resources listed in the data center through the Purchase menu under GPU Resources:
You can select the GPU you need and the length of time you expect to use it. Click on the purchase and you can complete the payment via Alipay or WeChat.
After payment, you can see the specifications of the purchased GPU on the Purchased page.
Start the GPU’s services
Once you have purchased the GPU service, you can see the list of GPU instances you can use from the Purchased menu. In the drop-down list, by clicking on the Operation drop-down menu on the right, you can launch a pre-prepared template service for you at the GPU instance. You can use these template services to further familiarize yourself with AI training and inference (usage) scenarios, so we have divided the templates into the following broad categories:
Note that when you start the GPU, you may run out of stock. Don’t worry, this may be due to too many current users, you can try again later.
Reasoning Class (Usage Class) - Entry Level:
Stable Diffusion WebUI & Jupyter NoteBook： This template allows you to quickly start an open source service for drawing based on cue words. This service also comes with a plugin for ControlNet and some plugins to facilitate fine-tuning the service. This image also contains a Jupyter Notebook service that you can rely on to complete file uploads and downloads of less than 100M in size.
Stable Diffusion WebUI With 3D to 2D Plugin& Jupyter NoteBook： This template adds a Plugin for converting 3D videos to 2D videos. This implementation is also based on Stable Diffusion, but with more plugins and some special SD settings and implementation steps. You can see how to do this in the video below.
With this video you can quickly learn how to use this template.
ChatGLM with LangChain： ChatGLM is a Chinese open source model with 6 billion parameters, in this template is integrated with the knowledge base Q&A function, which means that you can upload a document, either PDF or Word to this service to build a knowledge base. Then, based on this knowledge base, you can ask the corresponding questions and answers. This is especially suitable for customer bot scenarios.
One-click deepfake (face swap)： This is a small AI application, this template allows you to do the AI process of changing the face of a person in a video by uploading an avatar image and a video of the person. You just need to follow the step-by-step instructions in Jupyter Notebook to complete it.
With this video you can quickly learn how to use this template.
Learning category - Primary:
Jupyter Notebook： Jupyter Notebook is the most basic and commonly used software in machine learning, you can learn the Python language in this service. This template provides the most basic content and services for machine learning.
Jupyter Notebook for Pytorch：This is also a Jupyter Notebook service, but in this service pre-installed open source framework for machine learning Pytorch 2.0.1, you can base on this template to get exposed to basic concepts in the field of machine learning and do experiments to facilitate your familiarity with understanding linear regression, gradient descent, backpropagation and neural networks in the field of machine learning basic concepts.
Jupyter Notebook for Tensorflow：This is also a Jupyter Notebook service, but the open source framework for machine learning, Tensorflow 2.12.0, is pre-installed in this service. You can use this template to get exposed to basic concepts in the field of machine learning and do experiments to facilitate your familiarity with linear regression, gradient descent, backpropagation and neural networks in the field of machine learning basic concepts.
Training category - Intermediate:
Convert any voice to the target one：With this Template, you can convert any voice into an AI model of the target voice. This means you can train Jay Chou’s voice and sing Tsai Chin’s song by this Template. All you have to do is to prepare the material for your voice and follow Jupyter’s step-by-step instructions.
With this video you can quickly learn how to use this template.
This training requires more resources and it is recommended to ensure that the inference is trained in an environment with 24GB of video memory and 32GB of RAM.
Finetune a 3B Large Language Model with Lit-Parrot script： This Template will take you step-by-step through fine-tuning an open-source large model Stable-LM-3B using scripts provided by Lit-Parrot. Make it more comprehensible in Chinese.
With this video you can quickly learn how to use this template.
This training takes a long time, make sure you have enough time quota to complete the training.
GPU Storage
We know that in many machine learning and training scenarios, users will upload or download large amounts of data. The GPU service we provide is essentially a stateless service. Once the instance is down or restarted, all the data will be lost. That’s why we provide GPU storage service.
Once you purchase a certain capacity of storage, you can mount your storage to the specified directory when you start the GPU service. We also provide an interface for storage access. You can upload or download your data by clicking Open Web Console.
Stay tuned for new features!
We also understand that the basic templates currently integrated with the GPU service should fall far short of the needs of the majority of AI enthusiasts. For this reason, we will provide the ability to customize templates in the future. At that time, users can customize their templates and submit them to the platform. Once a template is submitted, it can be applied to GPU resources under the user’s tenant. Custom templates are usually in the form of Docker images. When submitting, users need to specify the relevant information, such as the service port information to listen to.
In addition, we will try to find some better GPU resources in the future. We will add some multi-GPU resources with high memory and high FLOPs to provide professional “alchemists” with professional “alchemy” needs.
Stay tuned for these exciting updates, and thank you for choosing MegaEase Cloud!
If you have any feedback, please feel free to contact us: [email protected]