verl-mint¶

verl-mint is the open-source edition of the MinT training runtime, built on veRL and Ray. It exposes the MinT training workflow at the API boundary, so clients keep familiar session, checkpoint, sampler, and future semantics while moving execution onto their own GPUs, Ray or veRL cluster, and artifact storage.

The same SDK script that runs against the MinT SaaS runs unchanged against a local verl-mint deployment.

Algorithms¶

Algorithm	Status	Notes
SFT	Available	LoRA SFT, checkpoint save/load, resume, sampler handoff
GRPO	Available	Group Relative Policy Optimization; group-relative advantages without a value critic
DPO	Planned	Preference optimization track

Supported models¶

Family	Size	Status
Qwen	0.6B	Available
Qwen	4B	Available
Qwen	30B	Available
Qwen	8B	Planned

Install¶

git clone https://github.com/verl-project/verl-mint.git
cd verl-mint
uv sync --extra smoke

The smoke extra installs the official MinT SDK (mindlab-toolkit) from the MindLab repository.

Prerequisites¶

verl-mint assumes the execution environment already exists; it does not provision Ray, GPUs, or worker images.

Prerequisite	Requirement
Ray	`ray[default]>=2.46.0,<3` on the API node and every worker

For the full setup guide, algorithm details, and cluster configuration, see the verl-mint repository.