With Blueprint, you only pay for what you use, down to the minute.
Interact with already deployed open-source models like Stable Diffusion and Whisper.
IncludesStable Diffusion, Whisper, FLAN-T5, and more Unlimited API calls Web IDE
Fine-tune open source models on custom data with ready-to-use fine-tuning scripts.
IncludesFoundation models Ready-to-use training scripts Programmatic fine-tuning API Dozens of customizable parameters
One-click deployment of your fine-tuned models on serverless GPUs.
IncludesAuto-scaling replicas Fast scale up from zero Custom idle scale down time Serverless functions
How much does it cost to fine-tune a model?
We charge for fine-tuning run time by the minute, so the cost to fine-tune a model depends on the model and the size of your custom training dataset. As an example, it takes about 30 minutes and therefore costs around $1.00 to fine-tune Dreambooth on 15 images.
Can I try Blueprint for free?
Invoking and building on top of already deployed open source models is always free. You also get 2 hours of GPU credits which you can use to try fine-tuning and model serving for free. These credits will automatically appear in your Blueprint account when you sign up.
How does billing work?
When your free credits run out, you’ll be asked to add a credit card to your account. We bill for fine-tuning run time and model serving time, by the minute. At the end of each month, we’ll charge you for your total usage throughout that month.
How does autoscaling work?
Fine-tuned models served on Blueprint will automatically scale up to a maximum of three replicas, based on traffic to the model. If your use-case requires a larger maximum, just talk to us and we can help you out.
Is there a cold start?
Cold start time will vary based on your model. As an example, fine-tuned Stable Diffusion will typically cold start in 30-40 seconds.
What is the idle scale down time?
By default, your fine-tuned model will scale down to zero after 30 minutes of inactivity. You also have the option to customize this idle scale down time for each model deployed on Blueprint.