Reference

Licensing

What license Meep² carries, what licenses the packages it installs carry, and what the other training engines' files say.

TL;DR

Meep² is Apache-2.0, and the packages it installs are permissively licensed: Apache-2.0, MIT or BSD, and MPL-2.0 for certifi and tqdm, a license that applies file by file and lets a product ship them unchanged. PyTorch and the CUDA libraries come from your environment under their own licenses. For a person running their own scripts, the license of an engine changes nothing; for a company shipping one inside a product, it does.

Meep² and what it installs

Every license was read in the repository’s license file at the version named in Meep²'s pyproject.toml.

Package License Where

Meep²

Apache-2.0

LICENSE; the vendored Axolotl kernels are credited in NOTICE

Transformers

Apache-2.0

https://github.com/huggingface/transformers/blob/main/LICENSE

PEFT

Apache-2.0

https://github.com/huggingface/peft/blob/main/LICENSE

TRL

Apache-2.0

https://github.com/huggingface/trl/blob/main/LICENSE

datasets

Apache-2.0

https://github.com/huggingface/datasets/blob/main/LICENSE

Axolotl’s LoRA kernels (vendored)

Apache-2.0

https://github.com/axolotl-ai-cloud/axolotl/blob/main/LICENSE

Liger Kernel

BSD-2-Clause

https://github.com/linkedin/Liger-Kernel/blob/main/LICENSE

bitsandbytes

MIT

https://github.com/bitsandbytes-foundation/bitsandbytes/blob/main/LICENSE

FlashAttention

BSD-3-Clause

https://github.com/Dao-AILab/flash-attention/blob/main/LICENSE

flash-linear-attention

MIT

https://github.com/fla-org/flash-linear-attention/blob/main/LICENSE

Triton

MIT

https://github.com/triton-lang/triton/blob/main/LICENSE

certifi (installed by requests)

MPL-2.0

https://github.com/certifi/python-certifi/blob/master/LICENSE

tqdm (installed by Transformers and datasets)

MPL-2.0 and MIT

https://github.com/tqdm/tqdm/blob/master/LICENCE

PyTorch

BSD-style

https://github.com/pytorch/pytorch/blob/main/LICENSE

NVIDIA CUDA libraries (from NVIDIA’s container, or the wheels pip installs with PyTorch)

NVIDIA’s license

https://docs.nvidia.com/cuda/eula/

vLLM (optional, for GRPO)

Apache-2.0

https://github.com/vllm-project/vllm/blob/main/LICENSE

Other training engines

As their license files read on 25 October 2026; a license can change, so read the files before relying on this page.

Unsloth. The package unsloth is Apache-2.0, and its README says so: "the core Unsloth package remains licensed under Apache 2.0, while certain optional components, such as the Unsloth Studio UI, are licensed under AGPL-3.0". A training with Unsloth also runs on unsloth_zoo, which unsloth requires and imports on import: its license file is the GNU Lesser General Public License v3.0 (https://github.com/unslothai/unsloth-zoo/blob/main/LICENSE). The LGPL is a copyleft license: whoever distributes a product that includes unsloth_zoo must offer its source with any modification and let the user replace that part. The wheel on PyPI also bundles the Studio UI: unsloth 2026.10.7 holds 1,972 files under studio/, with studio/LICENSE.AGPL-3.0, and 29 under unsloth_cli/, which the repository’s license file marks AGPL-3.0 as well, while the wheel’s metadata declares Apache-2.0 alone. GitHub issue unslothai/unsloth#8901 reported this on 2026-08-15 and was closed on 2026-10-11 without a comment; the wheel checked here is from after that date. unsloth_zoo declares no license in its PyPI metadata. So pip shows none of this.

What it means in practice:

  • Running Unsloth yourself, on your own machine or in a notebook: nothing to do. The LGPL and the AGPL bind the one who distributes, not the one who uses.

  • Shipping a product with Unsloth inside: the LGPL’s conditions apply to the unsloth_zoo part, and the AGPL’s to the Studio UI and unsloth_cli, which the wheel includes.

Meep² exists because Thinkube needed an engine with no such part; see Overview.