Skip to content

areal-project/AReaL-MinT

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

AReaL × MinT

AReaL-MinT

Open-source, self-hosted MinT training runtime on AReaL.

MinT SDK . Documentation . Quickstart . MinT Console . Cookbook . AReaL

Overview

MinT is RL training infrastructure for LLMs: a script on a CPU-only machine defines the data, loss, and RL environment, and MinT runs the computation across a GPU cluster. Switching models is a one-string change.

AReaL-MinT is an open-source, self-hosted MinT runtime: it exposes the same HTTP API as MinT but runs the work on AReaL (FSDP2 training) and SGLang (inference). The same SDK script runs unchanged against a local deployment.

Under the hood, a FastAPI control plane dispatches each request to its backend: training to AReaL's FSDP2 engine, inference to SGLang with LoRA hot-reload. Every write returns a future the client polls. See Architecture for the request path.

Tinker compatibility. MinT is compatible with the Tinker client surface. Existing Tinker code runs against AReaL-MinT unchanged by aliasing the import: import mint as tinker.

Support matrix

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

Models

Model family Size Status GPU layout
Qwen3.6 27B (Dense) Available 4 FSDP actor + 4 SGLang rollout
Qwen3.6 35B-A3B (MoE) Available 4 FSDP actor + 4 SGLang rollout

Quick start

Prerequisites

Requirement Version
GPU 8× NVIDIA (4 training + 4 inference)
Python >=3.11,<3.13
torch >=2.4 with CUDA
AReaL main
SGLang 0.5.10.post1
transformers >=5.0,<=5.3

Install

git clone https://github.com/areal-project/AReaL-MinT.git
cd AReaL-MinT

# Create venv inheriting system packages (torch, sglang, areal, transformers)
python3.12 -m venv --system-site-packages .venv
source .venv/bin/activate

# Install AReaL-MinT
pip install -e '.[gpu]' --no-deps
pip install pyyaml swanlab litellm openai-agents math-verify \
    colorlog tenacity uvloop aiohttp aiofiles httpx requests flask \
    fastapi pydantic uvicorn

For a full install from scratch with uv: uv sync --extra gpu --extra system.

Launch

export PATH=$PWD/scripts:$PATH
mintctl start 27b       # or: mintctl start 35b

Model loads in ~3 minutes. Check with mintctl status, view logs with mintctl logs -f.

Verify

bash examples/smoke_e2e.sh

Runs SFT → GRPO → RL convergence end-to-end (~15 min for 27B).

Architecture

MinT SDK ──HTTP──▶ AReaL-MinT (FastAPI control plane)
                        ├── service / router        session lifecycle & dispatch
                        ├── future_store            async result polling
                        ├── ArealTrainingBackend    FSDP2 training (SFT | GRPO)
                        └── ArealInferenceBackend   SGLang inference + LoRA hot-reload

Training requests go to AReaL's FSDP2 engine (4 GPU). Inference and sampling go to SGLang with LoRA hot-reload (4 GPU). All write operations return a future the client polls — same semantics as hosted MinT.

Usage

See the Developer Guide for a full walkthrough of SFT, GRPO, and multi-step RL training patterns.

Example Coverage
smoke_sft.py SFT train → sampling → checkpoint resume
smoke_grpo.py Full GRPO cycle: sample → rewards → advantages → train
smoke_rl_convergence.py Multi-step RL, asserts reward improvement

More examples and training recipes: mint-cookbook.

Citation

@misc{mindlab2026mint,
  title = {MinT: Managed Infrastructure for Training and Serving Millions of LLMs},
  author = {{Mind Lab}},
  year = {2026},
  eprint = {2605.13779},
  archivePrefix = {arXiv},
  primaryClass = {cs.LG},
  doi = {10.48550/arXiv.2605.13779},
  url = {https://arxiv.org/abs/2605.13779},
}

About

Open MinT training runtime on AReaL

Resources

License

Stars

24 stars

Watchers

1 watching

Forks

Packages

 
 
 

Contributors