Cumulus
The Most Liquid GPU Cloud
Infrastructure that adapts to your workload. Scale compute instantly, pay only for what you use.
Founded by alumni from
Process
How It Works
example.py
# Initialize client
client = CumulusClient("https://api.cumuluslabs.io")
# Run training with constraints
result = client.run(
func=finetune_llama2_7b,
budget="5.00",
optimization="time",
params=[model_config, dataset_path, num_epochs],
requirements=["torch", "transformers", "accelerate"]
)01
Submit
Define your workload with constraints—budget, deadline, or optimization target.
02
Profile
Our system analyzes requirements and identifies optimal resource allocation.
03
Allocate
Jobs run on shared GPUs, scaling across hosts. Switch providers for best prices.
04
Pay
Only pay for the compute you actually use. No idle costs, no overprovisioning.