Model Jobs
Each job type, from building the architecture to exporting for deployment, is tailored to streamline the workflow, reduce errors, and optimize performance for both high-demand and routine modeling tasks. They enable users to handle complex model operations with minimal coding, offering a user-friendly approach to sophisticated machine learning tasks.
Allowed operations that users can perform on the platform are:
- What is a Model Build Job?
- Submit a Build Job
- What is a Model Export Job?
- Submit an Export Job
- Terminate Job
- Delete Job
What is a Model Build Job?
The Model Build Job is a robust feature designed for constructing and testing machine learning model architectures.It validates models against various hardware requirements and architectural constraints, providing a detailed report of memory and CPU needs along with errors if any.
Users can choose the computational environment, such as shared or private virtual machines, to optimize performance and security.
Submit a Build Job
- From the model design canvas, go to Build Model → Submit Build Job.
- You will be shown the requirement data like Total Memory, Total Peak Memory along with the model design layers/nodes information.
- Next, select the virtual machine as per the requirement.
- Review the virtual machine information and click on Submit Build Job
- To view the build job information, go to Model Jobs from the header menu and click on the job you want to view. You will have the entire job information like Job Type, Progress, Status, Job Result, VM Configuration, Design Quick View, and a lot more organized in a single view.
You will have to create a model design first before submitting a build job. It is a good practice to run quick build before submitting a build job to validate any base level errors in the model design.
What is a Model Export Job?
The Model Export Job facilitates the seamless transition from model development to deployment. It allows users to export their trained models in various formats, including framework-specific code or weight files.
This feature also integrates with Scalifi Ai's Data Ingestion service to support a range of connectors, enabling efficient deployment across diverse environments like databases, cloud storage, or API endpoints.
Submit an Export Job
- Scalifi Ai currently supports model export as a python script or jupyter notebook. You can export the model design including both options.
- Make sure to Save and Build the model design at least once before exporting it.
- You need to have at least one file type connector configured with appropriate scope authorization to export the model, for example: AWS S3 Connector.
- Once the model design is ready, you can export the model design, by clicking on the Export button.
- Next, Save and Build the model design → Select source connector → Select target path.
- Select the export options and virtual machine (if needed) in Configure Export Options.
- Finally, review the export configuration and click on Export Model. You will find the exported file on the path you have chosen in the connector.
Terminate Job
- There are two ways to terminate a job.
- Go to the Model Jobs section from the header menu, select the job you want to delete → go to Actions → Terminate.
- From the model job page, select Action → Terminate
- You can submit termination request for only one job at a time.
- Termination is a dependent action. You will get proper errors while deleting a job if it is in active state.
Delete Job
- There are two ways to delete a job.
- Go to the Model Jobs section from the header menu, select the job you want to delete → go to Actions → Delete.
- From the model job page, select Action → Delete.
- You can submit deletion request for only one job at a time.
- Deletion is a dependent action. You will get proper errors while deleting a job if it is in active state.
