”Run a quick benchmark comparing PyTorch and JAX matmul performance on an RTX 4090.”
The agent handles everything end-to-end:create_instance: deploysbench-runwith RTX 4090instance_status: waits until runningssh_execute:pip install torch jax[cuda12]file_upload: uploads the benchmark scriptssh_execute:python /root/benchmark.pydelete_instance: terminates immediately
Benchmark results (RTX 4090, 4096x4096 matmul, 1000 iterations):Both frameworks perform nearly identically on this workload. JAX is marginally faster (~1.3%) due to XLA compilation. Billed time: 20 minutes on the RTX 4090, by the second. Instance terminated. Nothing left running.
”Test if my fine-tuned model loads correctly. Deploy a GPU, load it, run 5 prompts, and shut down.”
The agent runs the validation:create_instance: deploysmodel-testwith RTX 4090instance_status: waits until runningssh_execute:pip install torch transformersfile_upload: uploadstest_model.pywith 5 test promptsssh_execute:python /root/test_model.pybilling_usage: checks costdelete_instance: terminates
Model validation results:All 5 test prompts passed. Model loads in 8.2 seconds, inference at ~45 tok/s. Billed time: 10 minutes. Instance terminated.
”I want to try 3 different LoRA ranks and see which gives the best eval loss. Quick and dirty.”
The agent runs a fast comparison:create_instance: deployslora-testwith A100 80GBinstance_status: waits until runningssh_execute: installs dependenciesfile_upload: uploads training script and small eval datasetssh_execute: runs with--lora-rank 8(500 steps)ssh_execute: runs with--lora-rank 16(500 steps)ssh_execute: runs with--lora-rank 32(500 steps)delete_instance: terminates
LoRA rank comparison (500 steps each, Llama 8B, 1K sample dataset):Rank 16 wins. It achieves the lowest eval loss: rank 32 doesn’t improve further and trains 40% slower. The extra parameters at rank 32 are likely overfitting on this small dataset. Billed time: 16 minutes on the A100. Instance terminated.