# NOTICE

This repository contains **GGUF conversions** of the model
**NetEase Youdao Confucius4-R2T2** (https://huggingface.co/netease-youdao/Confucius4-R2T2,
https://github.com/netease-youdao/Confucius4-R2T2), produced by quantizing the
published `model.safetensors` checkpoint using an asymmetric per-tensor quantization recipe
researched and developed in the [transcribe.cpp](https://github.com/NairoDorian/transcribe.cpp)
project, with inspiration and tooling from [audio.cpp](https://github.com/0xShug0/audio.cpp)
and [davidxifeng/Confucius4-R2T2-gguf](https://huggingface.co/davidxifeng/Confucius4-R2T2-gguf).

## Derivative work

Quantized weights are a "Derivative Work" under the NetEase Youdao Model Use
License Agreement (section 1.5(iii)). These files are distributed under that
same agreement; the full text is in `LICENSE` (English) and `LICENSE_zh`
(Chinese) in this repository. Downstream recipients must comply with it.

Required statement, per section 4.1(a):

> Any modifications made to the original model in this Derivative Work are not
> endorsed, warranted, or guaranteed by the original right-holder of the
> original model, and the original right-holder disclaims all liability related
> to this Derivative Work.

中文：

> 本衍生品中对原始模型所做的任何修改，均未获得原始模型权利人的认可、保证或担保，
> 原始模型权利人对与本衍生品相关的所有责任予以免责。

## Attribution and Sources

* **Original model:** Confucius4-R2T2 — NetEase Youdao (copyright NetEase Youdao).
  * Hugging Face: https://huggingface.co/netease-youdao/Confucius4-R2T2
  * Repository: https://github.com/netease-youdao/Confucius4-R2T2
  * Demo: https://r2t2.youdao.com/demo
* **Base architecture:** Qwen3-ASR-1.7B (`Qwen/Qwen3-ASR-1.7B`).
* **Upstream GGUF baseline & inspiration:** David Xi Feng ([davidxifeng/Confucius4-R2T2-gguf](https://huggingface.co/davidxifeng/Confucius4-R2T2-gguf)) and the audio.cpp project (https://github.com/0xShug0/audio.cpp).
* **Asymmetric Q4_K_M Quantization Recipe (Arm M):** Engineered and benchmarked for [transcribe.cpp](https://github.com/NairoDorian/transcribe.cpp) to solve the low-bit quantization barrier and enable full multilingual stability at 1.187 GB.

## License reminders

* Section 2.2: if you or your affiliates had more than 100 million monthly
  active users in the preceding calendar month, or annual revenue above
  RMB 1 billion in the preceding calendar year, you must request a separate
  commercial license from NetEase Youdao.
* Section 3.4(c): the model and its derivatives may not be used to improve any
  other AI model, except Confucius4-R2T2 itself, its derivatives, or
  non-commercial AI models.
* Section 4.2: prohibited high-risk uses include medical diagnosis, autonomous
  driving, military applications, critical-infrastructure control, large-scale
  biometric surveillance, and automated decision-making such as credit or
  employment evaluation.
* Section 5.2: no trademark rights are granted; the names are used only to
  describe the origin of this Derivative Work.