In order to avoid the extreme unexpected charges that are common with other cloud providers, trainML uses a credit-based billing system that ensures you only spend what you planned. Credit purchases are referred to as "payments" and the consumption of credits by using platform resources are referred to as "charges."
Credits can be purchased for $1.00 USD each. No resources can be provisioned on the platform without a positive credit balance. Prices for all resources are denominated in credits per time interval. Credits do not expire. Your credit balance can be viewed at all times in the top right area of the header toolbar.
To manage payment methods, purchase credits, or view your payment history, navigate to the
Billing section using the sidebar. Alternatively, you can click
Credits in the top right area of the screen.
trainML payments are powered by Stripe. trainML itself cannot access or store your payment card details.
Payment Methods section displays all configured payment method. If you have multiple payment methods, the primary payment method will be shown with a label of
Primary. To add a new payment method, click
Add Payment Method. Input the billing details of the payment method. On the next page, enter your credit card information.
Note: Only US credit cards are currently supported.
You can remove a payment method by clicking the
Remove button. This will automatically make one of your remaining payment methods the primary method.
When you return to the payments page, click
Set Primary on your new payment method to set it as the payment method to use for the new transaction. Once it says Primary under the payment method, type the amount of credits you want to purchase into the
Amount of Credits field on the left and click
Buy. Confirm your payment on the popup by clicking
Payments made through the
Purchase Credits form will have a type of
Manual in the
Payment History table.
By default, trainML will not bill your credit card unless you manually purchase credits. If you run out of credits while a job is running, the job will be stopped. To prevent this, you can enabled Automatic Top-ups by clicking the
Set Automatic Top-Up button. Check the
Enable automatic Top-up checkbox to enable the form. By default, every time an automatic top-up is performed, you will receive an email notification. If you wish to disable this, uncheck the
Send me an email notification when a top-up occurs box. The
Top-up Credits Threshold is the level of credits your account must dip below before an automatic top-up is triggered and the
Top-up Credits Amount is the amount of credits to purchase once that threshold is passed. For example, if you set the
Top-up Credits Threshold to
10 and the
Top-up Credits Amount to
30 and you have a job running that consumes 1.5 credits per hour with a current account balance of 11, when the job bills you for an additional hour your credits drop to will drop to 9.5. This is below the credits threshold of 10 and will therefore trigger a top-up. Your default payment method will automatically be charged for 30 credits (your
Top-up Credits Amount) and, if the payment is successful, your credit balance will be 39.5.
Payments made through the automatic top-up process will have a type of
Automated in the
Payment History table.
Consolidated billing allows one account (the billing account) to act as the source of credits for multiple user accounts. The billing account becomes the only account with a credit balance, associated payment methods, and automatic-topup settings and is the only account that can view the payment history. Whenever a user account provisions resources that require credits, the credits are deducted from the billing account's credit balance. The billing account can also create jobs and use its own credits, but it does not have access to the jobs, models, or datasets of any other user accounts. The billing link can be terminated by either account (billing or user) at any time. If the billing link is terminated while jobs are active or storage usages is above the free tier, the unlinked user account will need to add credits to keep the resources active.
Billing linking can only be initiated from the user account that wants to use the credits of another account. To link your account to another, enter the email address of that user's trainML account and click
Send Request. The request will remain pending until the recipient accepts it or you cancel the request.
If you have paid credits still available in your account, those credits will be transferred to the billing account once they accept your request. Any non-paid (e.g. coupon) credits will be forfeited. This process is not reversible. Be sure you have the correct email address.
After a request is made, when the billing account goes to the
Consolidated Billing section on the billing page, they will see
Pending Consolidated Billing Requests with the name, email address, and profile picture of the requestor. If they approve the request, that user account will move to the
Active Consolidated Billing Links section. At this point, all activity for that user's account will use the credits of the consolidated billing account.
The billing account can remove a linked user at any time by clicking the
Remove button on the user card in the
Active Consolidated Billing Links section. The effect will be immediate and will cause all jobs to stop when they attempt to reserve more credits unless the user configures a payment method and purchase credits. A user account linked to another billing account can also remove the link by clicking the
Deactivate button on the billing page.
trainML charges credits for the use of platform resources. The two categories of resources are Compute and Storage.
Compute charges are incurred by a running jobs (Notebooks, Training Jobs, Inference Jobs, or Endpoints). The amount charged is based on the credits per hour of the GPU type selected and multiplied by the number of GPUs requested and number of workers. For example, if a GPU Type is 0.1 credits per hour, and you start a 10 worker training job with 2 GPUs per worker, the hourly charge would be 2 credits (0.1 * 10 * 2). The credits per hour each GPU type costs can vary based on the supply and demand of that GPU type. Once you create a job, your credits per hour rate is locked in as long as the job is running. To see the most up-to-date prices, login to the platform and click
Start a Notebook or
Create a Training Job on the Home page.
One hour's worth of compute credits are deducted when a job is started, and continue to be deducted each hour as long as the job is running. When the job stops, the deducted credits for the remainder of the hour are refunded. If you are running a job with multiple workers, each worker is billed independently. For example, if you have 3 workers with 1 GPU each at 0.1 credits per hour and worker 1 runs for 30 minutes, worker 2 for 60 minutes and worker 3 for 150 minutes, the total charge will be 4 credits (0.1 * 0.5 + 0.1 * 1 + 0.1 * 2.5).
If you do not have enough credits to pay for the next hour of compute for a job worker, it will automatically stop. After your credit balance is increased, you can restart the job.
Storage charges are incurred by creating private datasets and allocating disk space to job workers. The first 50 GB of storage is free each month, the storage charge is 0.20 credits per GB per Month for any storage in excess of 50 GB. trainML Private datasets and models are charged based on their actual size of the once created. The storage charge for job workers is based on the amount of storage provisioned for each worker (not used). The storage charge for notebooks continue for as long as they exist, not just when they are running. For example, if you have one private dataset of 50 GB and two notebooks that each have a 20 GB disk size for an entire month, the total monthly charge would be 8 credits ( (50 + 2 * 20 - 50) * 0.2 ). Training and inference job storage charges start when the job is created and stop when the job finishes.
Credit charges for storage occur daily and are retrospective. If you no longer have enough credits to pay the storage charges, your datasets and workers may be purged.