# Third-party notices

This repository contains a transformed subset of the weights published in
[`deepseek-ai/DeepSeek-V4-Flash-0731`](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731)
at commit `9e165c30e2704aec5d9d593cce3eebd58bbef1cb`.

DeepSeek publishes that repository under the MIT License. Its copyright and
permission notice are available in the pinned checkpoint's
[upstream license](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731/blob/9e165c30e2704aec5d9d593cce3eebd58bbef1cb/LICENSE).

The standalone conversion and verification scripts are newly authored for
this repository and published under the repository's declared MIT license.

The conversion scripts implement publicly documented numeric and file formats.
The Q2_K encoder in `scripts/convert.py` is a Python port of the unweighted
`ds4q_write_q2_k_block_ref` path and its numeric helpers from
[`antirez/ds4` at commit `453a5faa899759d0edf7729d595c4c39cddfcfca`](https://github.com/antirez/ds4/blob/453a5faa899759d0edf7729d595c4c39cddfcfca/gguf-tools/quants.c).
That source identifies its block layout and search procedure as derived from
the MIT-licensed GGML/llama.cpp quantizers.

The applicable DS4/GGML MIT notice is:

> Copyright (c) 2026 The ds4.c authors  
> Copyright (c) 2023-2026 The ggml authors
>
> Permission is hereby granted, free of charge, to any person obtaining a copy
> of this software and associated documentation files (the "Software"), to deal
> in the Software without restriction, including without limitation the rights
> to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
> copies of the Software, and to permit persons to whom the Software is
> furnished to do so, subject to the following conditions:
>
> The above copyright notice and this permission notice shall be included in
> all copies or substantial portions of the Software.
>
> THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
> IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
> FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
> AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
> LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
> OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
> SOFTWARE.

No PyTorch, llama.cpp, or GGUF Python package is required at build time.

Format references:

- [DeepSeek-V3 FP8 block scaling implementation](https://github.com/deepseek-ai/DeepSeek-V3/blob/main/inference/kernel.py)
- [OCP Microscaling Formats specification](https://www.opencompute.org/documents/ocp-microscaling-formats-mx-v1-0-spec-final-pdf)
- [GGUF format documentation](https://github.com/ggml-org/ggml/blob/master/docs/gguf.md)
- [DS4 reference quantizer](https://github.com/antirez/ds4/blob/453a5faa899759d0edf7729d595c4c39cddfcfca/gguf-tools/quants.c)
- [DeepSpec](https://github.com/deepseek-ai/DeepSpec)