tuneplane

Client for TunePlane. You get the tuneplane command and a Python package named tuneplane.

Use this to scaffold a lab, lock a post-training method, and submit jobs to a TunePlane console. The control plane is not on PyPI: it ships as a container image and a Helm chart. Training images do not need this whole package; the server injects a runner that carries the kernel only.

Install

uv tool install tuneplane
# or, without uv
pip install tuneplane

tp --help

uv tool install is recommended: it gives tuneplane its own environment, and it is the installation tp update can upgrade. Installed into a system Python, tp update refuses rather than risk disturbing other tools there.

Upgrade with tp update — it detects how tuneplane was installed and runs the right command, or names the one you should run when upgrading in place would break something.

Python 3.12 or newer. Import name is tuneplane, not tuneplane.

uv tool install "tuneplane[data]" adds HuggingFace datasets for local prepare scripts.

Start a lab

Do not clone the platform repo. This package writes a project for you:

tp init my-lab --yes
cd my-lab
tp login --server https://<your-console>
tp new my-grpo --method nemo-rl/grpo
tp validate my-grpo
tp submit my-grpo --profile h200:8
tp job logs

tp methods lists the catalog on the server you logged into. Methods look like nemo-rl/grpo, verl/sft, trl/kto, openrlhf/ppo, evalkit/benchmark, custom/custom.

Examples live in tuneplane-examples. Day-to-day work should be your own tp init tree.

Python

Catalog methods already forward framework logs. Custom train.sh jobs that want console curves call:

from tuneplane.report import init, log, finish

init()
log({"loss": 0.12}, step=1)
finish()

tuneplane.spec_to_env is the JobSpec to environment mapping used on both the client and the job. tuneplane.get_recipe / tuneplane.recipe_names read the bundled catalog.

What this package contains

• tuneplane CLI: init, login, experiments, submit, sweep, jobs, datasets, plugins, bench, serve
• JobSpec types and spec_to_env
• Recipe catalog and framework adapters (NeMo-RL, verl, TRL, OpenRLHF, evalkit, custom)
• Job-side reporter (tuneplane.report)

Cluster credentials stay on the server. The client keeps an access token under ~/.tuneplane/.

Docs and source

• Platform repo: https://github.com/tuneplane/tuneplane
• User docs: mintlify/ in that repo (English and 简体中文)