Plugger AI
Comment on page

Cold & Warm States

Welcome to the technical documentation for's AI models. In, AI models have two distinct states: Cold State and Warm State. These states are designed to optimize the utilization of AI models while keeping costs in check, simulating a serverless infrastructure. This documentation provides a detailed understanding of these states and how they function.

Cold State

In the Cold State, the AI model is not actively running but remains available for use. When you send a request to a model in the Cold State, the model needs to initialize before it can handle the request. The initialization time may vary depending on the specific model type, but on average, it takes approximately one minute. After the initialization process is complete, the model will be ready to process your request and provide a response.
Key Points:
  • AI models in Cold State are not actively running.
  • Initialization is required before handling a request.
  • Initialization time varies by model but typically takes around one minute.
  • After initialization, the model will generate the requested output.

Warm State

In the Warm State, the AI model has been previously used and remains active without the need for initialization. This state is designed to eliminate the initialization process, ensuring rapid response times for consecutive requests. However, there is an important aspect to consider: if the model does not receive a request within 3 minutes, it will automatically shut down and transition back to the Cold State.
Key Points:
  • AI models in Warm State are actively running and ready to process requests.
  • No initialization process is required in the Warm State.
  • Models will remain in the Warm State as long as they receive requests.
  • If a model remains idle without requests for 3 minutes, it will return to the Cold State.

Why We Use These States

At, we recognize that GPU instances can be expensive, making AI services less accessible to a wider audience. To address this challenge, we have developed a unique system that optimizes AI model utilization. By implementing Cold and Warm States, we can make AI accessible at a more affordable price point, allowing us to serve AI models that may not be available elsewhere within this cost range.
  • Cost-efficient utilization of GPU resources.
  • Reduced waiting time for AI model initialization in the Warm State.
  • Affordable AI services for a broader audience.
  • Access to a variety of AI models at competitive prices.

Conclusion's AI models employ Cold and Warm States to optimize resource utilization, reduce costs, and provide affordable access to AI capabilities. These states allow us to deliver efficient AI services, ensuring that our users can benefit from the power of AI without breaking the bank. Whether you require rapid responses or are mindful of cost constraints,'s AI model states have you covered.