Key concepts
Here is an introduction HyperAI Some important concepts mentioned in.
Computing power container Gear
Computing power container, In documents, it is often referred to as "container" , It can be simply understood as a small machine that provides specific resources. When creating a container, it is necessary to specify "Computing power" (CPU. Memory. Hard disk size), choice "image" (Can be understood as pre installed applications), binding "data" .
Each time the container "implement" Each will provide independent storage space and be bound to /hyperai/home catalogue. As its working directory. After the execution is completed. The content will also be saved.
Each execution can be created through a directory
/hyperai/input/input0/hyperai/input/input1/hyperai/input/input2/hyperai/input/input3/hyperai/input/input4
Bind other data.
For more information about data binding, please refer to Data binding.
Gear of "implement" Currently supports two methods:
- Python script
- Jupyter working space
among "Python script" Support users to upload their own code. Then specify the command to be executed. The container will execute the entry command specified by the user upon startup Python script. Suitable for one-time use. Not requiring frequent modifications Long task implement. For example, once lasting for a long time 12 An hour long model training task.
Jupyter A workspace is an interactive code execution tool, establish Jupyter After the workspace, we can perform real-time command execution and exploratory work on it. Suitable for debugging and building early models.
Jupyter Once the workspace is in operation, billing begins. Until the user voluntarily closes it. Even if the user is not using the computing resources within it. But due to the exclusive nature of computing power resources. It will still be charged normally.therefore. Once no longer in use Jupyter working space. Please close it promptly. Avoid unnecessary billing.
The same one "container" Below "implement" Considered to have close business connections, every time "Computing power container" of "implement" They will all be independently recorded.
But every time "implement" The environment is all isolated. That means every time there is "implement" The installation command executed in the middle cannot be retrieved after this execution is closed. Even though through "Continue to execute" Running up "implement" The environment is still a completely new one.
Data warehouse Data Warehouse
Users can create data warehouse Used for saving and reusing data. Currently divided into two types: Dataset and Model. Generally speaking, the size of datasets is quite large. It is very unrealistic to repeatedly upload code and data every time it runs. Therefore, a separate area is provided here for the management of the dataset.
Bind data warehouse
Creating "Computing power container" When, adopt "Bind data" You can bind the data warehouse to a container for use, at present "container" Supports the most 5 Data binding.
As shown in the following figure. When creating a container, you can choose a data warehouse or the one mentioned below "working directory" Bind to /hyperai/input/input0 /hyperai/input/input1 /hyperai/input/input2 /hyperai/input/input3 /hyperai/input/input4 /hyperai/home In any directory within. When creating a container. The corresponding data warehouse or working directory will be bound to the corresponding directory.
As shown in the above figure, hyperai/cifar-100 Image classification dataset/2 The data warehouse will be bound to the container /hyperai/input/input0 catalogue. After the container is running, it can be /hyperai/input/input0 Visited it.
User's "storage space "
User uploaded data warehouse, code. The content saved after the container is closed will be saved in the user's account "storage space " in. On the user's personal page, they can see the usage of their storage resources. Once the user storage exceeds the total amount, they will no longer be able to create containers. Upload data.
More information can be found at Resources and usageget.
The container's "working directory"
"working directory" It refers to each "implement" In the middle /hyperai/home catalogue. The data in this directory will be automatically saved after the container is closed. Its usage will also be reflected in the user's "storage space " in.in other wordsAll the content you want to save should be stored in this directory.
Default binding in container /hyperai/home of "working space" With "storage space " The relationship is as follows:
- The container's "working space" The capacity is
/hyperai/homeThe upper limit of storage capacity. When using this container. The total capacity of files stored in this directory cannot exceed this limit. - "working directory" The data will be persistently saved. Even after the container is closed, it will still be saved. So of course, this part of the data will also take up space "storage space " Capacity of.
Computing power
The definition of computing power lies in HyperAI How many resources are provided for container execution, at present HyperAI Mainly provides two types of computing power:
- CPU Computing power. Containing a certain number of CPU, adopt CPU Execute machine learning algorithms
- GPU Computing power. Include one or even multiple GPU. Can utilize deep learning frameworks GPU Version Acceleration for Training and Reasoning of Deep Learning Models
Each type of computing power provides an upper limit for memory and storage. Once the usage limit of memory or storage is exceeded during container operation, it will result in task failure.
Runtime environment
Runtime environmentAlso known as a mirror image. Provides the types of main machine learning algorithm libraries that containers rely on during execution.at present HyperAI Provided TensorFlow. PyTorch. MxNet. Darknet of CPU and GPU Different versions of standard libraries in different environments. More types of machine learning algorithm libraries will be provided in the future.
For more information, please refer to Runtime environment.