Skip to main content
Deploy a fleet of instances, run the same command on all of them, and monitor everything at once. Your AI agent handles the coordination.

”Deploy 4 RTX 4090 instances for a hyperparameter sweep.”

The agent provisions the fleet:
  1. list_gpu_types: confirms RTX 4090 availability and pricing
  2. create_instance: creates sweep-1 with RTX 4090
  3. create_instance: creates sweep-2 with RTX 4090
  4. create_instance: creates sweep-3 with RTX 4090
  5. create_instance: creates sweep-4 with RTX 4090
  6. instance_status: polls all four until running
Deployed 4 instances: All 4 are ready for commands, each billed at the RTX 4090’s live rate while it runs.

”Install dependencies on all of them and start training with different learning rates.”

The agent runs commands across the fleet:
  1. ssh_execute on each instance: pip install torch transformers datasets
  2. file_upload on each instance: uploads train.py
  3. ssh_execute on each instance: starts training with different args:
    • sweep-1: python train.py --lr 1e-4
    • sweep-2: python train.py --lr 5e-5
    • sweep-3: python train.py --lr 1e-5
    • sweep-4: python train.py --lr 5e-6
All 4 instances are running training with different learning rates: Check back in ~30 minutes for results, or ask me to monitor progress.

”Check training status on all instances and tear down any that are done.”

The agent checks each instance and cleans up:
  1. ssh_execute on each: tail -5 /root/train.log
  2. ssh_execute on each: nvidia-smi --query-gpu=utilization.gpu --format=csv,noheader
sweep-2 (lr=5e-5) has the best loss so far at 0.287. Terminating finished instances to save money.
  1. file_download on sweep-1 and sweep-2: downloads results
  2. delete_instance: terminates sweep-1 and sweep-2
Downloaded results from sweep-1 and sweep-2. Terminated both. 2 instances remain, still billing.

Tools used in this workflow