Introduction
For a quick start, please refer to the Getting Started guide.
Overview
Dat1.co is a serverless GPU platform that is designed to make real-time inference easy.
Here are some of the key features:
- Ultra-low cold start: We cut down typical cold start times from minutes to seconds. For a 40GB model, you can expect a cold start under 20 seconds. This not only makes the first request faster but also makes the platform more cost-effective.
- Managed GPU: We take care of the hardware management, auto-scaling, and high availability. You can focus on your model and the application.
- GPU availability: Due to lower cold start times, you can expect dozens of GPUs to be available at any given time.
Common Environment
Dat1 uses a common docker image for all the models. You can check the Dockerfile here.
If your model relies on a library that is not included in the standard Docker image, you can package it with your model using pickle or another serialization method. If you believe a library should be included in the image, please contact us at contact@dat1.co, and we will consider adding it.