How do I install the Areno Tune Capacity skill?

Run npx skillmds@latest add inclusionai/areno-tune-capacity in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Areno Tune Capacity safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Areno Tune Capacity?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Areno Tune Capacity free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Areno Tune Capacity?

inclusionai (@inclusionai) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Areno Tune Capacity by inclusionai — Agent Skill

Name: Areno Tune Capacity
Author: inclusionai

Areno Tune Capacity

Fit an AReno training or rollout workload to available GPUs by tuning TP, mini-batch, batch, sample count, and rollout concurrency. Use for OOM prevention, memory headroom, smoke-infer, smoke-train, or tune-params requests. Do not change semantic token limits unless requested.

inclusionai Updated Sep 17, 2026

File contents

Tune AReno Capacity

Inspect current train help and areno/cli/auto_tune.py. Validate relationships first:

python .agents/skills/areno-tune-capacity/scripts/check_capacity.py \
  --batch-size N --n-samples N --max-running-prompts N \
  --mini-bs N --world-size N --tp-size N

Workflow

Record GPU count/memory, model config, dtype, TP constraints, optimizer, and semantic token lengths.
Measure rollout with --smoke-infer when useful; it must allocate cache and capture decode graphs.
Measure train with --smoke-train; it skips rollout/prefill and uses the candidate microbatch.
Use --tune-params when requested. Keep peak memory at or below the requested fraction, never above 0.9 when selecting a default safety target.
Avoid excessive probes. Probe to make a decision, then confirm the chosen setting.
Run a bounded real workload if the overall user goal is a working task.

Preserve max_new_tokens and max_context_len. Read references/parameter-relations.md before adjusting multiple dimensions.

Frequently asked questions