Configuration: Fine-Tune Your Training Jobs

Trainwave’s flexible configuration system allows you to precisely define the environment and resources required for your machine learning jobs. By specifying parameters in your trainwave.toml file, you can optimize your training process for efficiency and cost-effectiveness.

How Configuration Works

Trainwave’s intelligent matching system analyzes your configuration parameters to find the best available resources that meet your needs. While you have the freedom to specify as many or as few parameters as you like, providing more specific requirements may narrow down the available options and potentially impact pricing.

Configuration Options

Here’s a comprehensive guide to the available configuration options:

Option	Required	Type	Description
`name`	True	String	The name of your job. This does not need to be unique.
`project`	True	String	The ID of the project this job belongs to (e.g., `p-jsefhsee`).
`setup_command`	True	String	The command to be executed first to set up your environment. This could involve installing dependencies or running a setup script.
`run_command`	True	String	The command to start your training process (e.g., `python train.py`). This can also point to a bash script.
`expires`	False	String	Automatically stop the job after a specified duration (e.g., `1h` for 1 hour, `10m` for 10 minutes, `1d` for 1 day). This helps prevent runaway costs.
`env_vars`	False	Object	Define environment variables for your job. You can set fixed values (e.g., `env_vars.ENV_VAR = "abc"`) or use interpolation to access your local environment variables (e.g., `env_vars.ENV_VAR = "${MY_LOCAL_ENV_VAR}"`).
`exclude_gitignore`	False	Bool	Set to `true` to prevent files and folders specified in your `.gitignore` file from being uploaded to the training environment. This can help reduce upload times and storage costs.
`exclude_regex`	False	String	Use a regular expression to exclude specific files or folders from being uploaded (e.g., `exclude_regex = "data.*"` to exclude any file or folder starting with “data”).
`image`	True	String	The Docker image to use for your job. See the Images documentation for available options.
`hdd_size_mb`	True	Integer	The required disk space in MB (maximum 500GB).
`memory_mb`	False	Integer	The minimum amount of RAM required for your job.
`cpus`	False	Integer	The minimum number of CPUs required.
`gpus`	False	Integer	The number of GPUs required.
`gpu_type`	False	String	The specific type of GPU required. Refer to the GPU Types documentation for available options.
`compliance_soc2`	False	Bool	Set to `true` if you require SOC 2 compliant data centers for your job.

Optimizing Your Configuration:

Start with essential parameters: Begin by defining the core requirements for your job (name, project, setup_command, run_command, image, hdd_size_mb).
Add constraints as needed: Introduce additional parameters like gpus, gpu_type, memory_mb, and cpus to fine-tune your resource allocation.
Use exclusion rules: Utilize exclude_gitignore or exclude_regex to avoid uploading unnecessary files and improve efficiency.
Consider expires for cost control: Set an expiration time for your jobs to prevent unexpected expenses.

By carefully configuring your training jobs, you can ensure optimal performance, efficient resource utilization, and predictable costs with Trainwave.

CLI GPU Types