Remote Run Over SSH FAQs

Question 1

How does it optimize the workspace synchronization?

Accepted Answer

The skill uses optimized rsync patterns with specific include/exclude rules. It only transfers the essential library files and the specific example folder you are working on, significantly reducing transfer times compared to syncing an entire repository.

Question 2

What does the Remote Run Over SSH skill do?

Accepted Answer

This skill allows Claude to automate the process of running machine learning workloads on remote servers. It synchronizes your local code to a GPU host via SSH, builds Docker images, executes training scripts, and tails logs back to your terminal for real-time monitoring.

Question 3

What are the technical prerequisites?

Accepted Answer

You need a configured SSH alias (mapping to your GPU host), a remote environment with NVIDIA GPUs and Docker (with GPU runtime), and approximately 15GB of free space in the remote /tmp directory for the workspace and Docker context.

Question 4

Can I retrieve training artifacts and logs automatically?

Accepted Answer

Yes. The skill includes procedures for capturing metrics and rsyncing artifacts like run logs, model checkpoints, and TensorBoard data from the remote host back to your local environment for documentation and reproducibility.

Question 5

When should I use this skill?

Accepted Answer

Use this skill for heavy training or evaluation tasks that require CUDA acceleration not available on your local machine, such as CIFAR10 speed runs, multimodal pipeline training, or large-scale generative AI fine-tuning.

Remote Run Over SSH

Remote Run Over SSH

Key Features

Use Cases

Key Features

Use Cases