rl-standard-launch-leonardo

Launch and manage offline SkyRL experiments on CINECA Leonardo via sbatch wrappers.

276|37|Updated Dec 5, 2025

Search Tags:#offline #grid #rl #multinode #sbatch #leonardo #skyrl

Ready to run

npx skills add https://github.com/open-thoughts/OpenThoughts-Agent --skill rl-standard-launch-leonardo

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: rl-standard-launch-leonardo
Source: https://github.com/open-thoughts/OpenThoughts-Agent/tree/main/.claude/skills/rl-standard-launch-leonardo
Command: npx skills add https://github.com/open-thoughts/OpenThoughts-Agent --skill rl-standard-launch-leonardo

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Launch, relaunch, or sweep standard SkyRL RL experiments on CINECA Leonardo in offline environments using sbatch wrappers, a writable apptainer sandbox, and uv venv, avoiding the traditional hpc.launch workflow.

Core Features & Use Cases

Supports GSM8K, MATH/aime datasets, OPD/distillation workflows, single- and multi-node RL runs.
Uses hpc/skyrl_yaml/leonardo wrappers with sandboxed execution and offline data staging.
Provides guardrails for no-internet environments and guidance for grid-cell experiments and performance profiling.

Quick Start

Submit the GSM8K canary job on Leonardo by running sbatch sbatch_gsm8k_canary.sh from the leonardo launcher directory.

Frequently Asked Questions about rl-standard-launch-leonardo

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I launch offline SkyRL experiments on CINECA Leonardo?▼

To launch offline SkyRL experiments on CINECA Leonardo, use sbatch wrappers under hpc/skyrl_yaml/leonardo with a configured uv venv and writable apptainer sandbox. This avoids the traditional hpc.launch workflow for single-node and multi-node runs.

Can I run multi-node reinforcement learning jobs on Leonardo without internet access?▼

Yes, you can run multi-node reinforcement learning jobs on Leonardo without internet access. The Skill provides guardrails for no-internet environments and uses offline data staging to support GSM8K, MATH/aime, and OPD workflows.

What is the best way to start a GSM8K canary job on Leonardo?▼

The best way to start a GSM8K canary job on Leonardo is to run sbatch sbatch_gsm8k_canary.sh from the leonardo launcher directory. This tests your offline SkyRL setup using the sbatch wrappers and uv venv.

Does the SkyRL launcher support OPD and distillation workflows on a multinode grid?▼

Yes, the SkyRL launcher supports OPD and distillation workflows on a multinode grid. It handles single-node and multi-node RL runs using sbatch wrappers and offline data staging on CINECA Leonardo.

Do I need a writable apptainer sandbox to run MATH/aime experiments on Leonardo?▼

Yes, you need a writable apptainer sandbox to run MATH/aime experiments on Leonardo. The sandboxed execution environment is required alongside sbatch wrappers, offline data staging, and a configured uv venv.

Why avoid the traditional hpc.launch workflow for offline SkyRL experiments?▼

You should avoid the traditional hpc.launch workflow for offline SkyRL experiments because it lacks the necessary guardrails for no-internet environments. The Skill uses sbatch wrappers and a writable apptainer sandbox to manage offline data staging directly.