Serverless Functions

how it will work

How it fits into GPU.ai

Serverless Functions will run your own Python code on GPU.ai without renting a machine or building a container. You add one decorator to a function and call it. GPU.ai starts a GPU machine for the call, runs the function on it, and sends the result back.

from gpuai_functions import function

@function(gpu="a100_80gb")
def gpu_name():
    import torch
    return torch.cuda.get_device_name()

gpu_name.remote()  # wait for result
gpu_name.spawn()   # run in background

• One decorator

  Mark a function with @function and name a GPU type, or leave it out and take the cheapest GPU that fits.

• Wait for it, or run it in the background

  .remote() returns the function's result. .spawn() returns at once with a handle you can check later.

• Warm between calls

  A machine stays ready for an idle timeout you choose, so the next call skips start up. Then it is released.

• Billed for machine time

  You will pay for the time the machine exists, on the same GPU.ai account you use for instances.