hw-whistleblower/true-story-of-pangu

★ 11,411Audience · generalComplexity · 1/5Setup · easy

Mindmap

mindmap
  root((True Story of Pangu))
    What it claims
      Model copied from Qwen
      Leadership was aware
      Internal suppression
    Key entities
      Huawei Noah Ark Lab
      Alibaba Qwen model
      Pangu 135B V2
    Context
      AI development China
      Whistleblower account
    Content type
      Chinese text document
      No code or software

mindmap root((True Story of Pangu)) What it claims Model copied from Qwen Leadership was aware Internal suppression Key entities Huawei Noah Ark Lab Alibaba Qwen model Pangu 135B V2 Context AI development China Whistleblower account Content type Chinese text document No code or software

Click or tap to explore — scroll the page freely

Things people build with this

USE CASE 1

Read as a primary source document about AI development practices inside large Chinese technology companies.

USE CASE 2

Use as a reference when researching the Huawei Pangu versus Alibaba Qwen model controversy.

USE CASE 3

Understand how large AI models can be derived from existing ones and why that raises ethical questions.

Getting it running

Difficulty · easy Time to first run · 5min

In plain English

True Story of Pangu is not a software project. It is a public disclosure document written by a self-described employee of Huawei's Noah's Ark Lab, the research division that developed the Pangu family of large language models (AI systems similar in concept to ChatGPT). The repository contains a single lengthy text in Chinese laying out allegations about how those models were built. The author describes the working conditions during the development period: long stretches away from home during concentrated work sessions in Suzhou, heavy delivery pressure from business units, and an internal culture where research had shifted toward product delivery. The core allegation is that a key version of the Pangu model, labeled 135B V2, was not trained from scratch as publicly claimed but was instead derived from Qwen, a competing model released by Alibaba. The author states that Huawei's internal team fine-tuned Qwen's weights, added layers, and rebranded the result, with multiple levels of leadership reportedly aware but choosing not to intervene. The account includes specific technical details intended to establish the author's credibility: naming senior researchers by name and title, describing internal organizational structure, and explaining model architecture decisions and their outcomes. Colleagues within the team reportedly referred to the resulting product informally as a portmanteau combining the names of the two models, a darkly humorous nickname reflecting their awareness of the situation. The document was posted publicly during a period of wider controversy in China around allegations that Huawei's Pangu model had copied Alibaba's Qwen model. The author expresses personal conflict about going public but states they could no longer stay silent after seeing internal attempts to suppress the information. This repository contains no code and no runnable software. It is a primary source document for anyone following the story of AI development practices inside large Chinese technology companies.

Copy-paste prompts

Prompt 1

Summarize the key allegations in the True Story of Pangu disclosure and explain what fine-tuning a model's weights means in plain English.

Prompt 2

Based on this whistleblower account about Huawei's Pangu model, what are the ethical implications of rebranding a fine-tuned open-source model as an original creation?

Prompt 3

Explain what it means technically to add layers to an existing AI model and rebrand it, and how researchers might detect if one model was derived from another.

Open on GitHub → Explain another repo

← hw-whistleblower on gitmyhub — every repo by this author, as a profile.

Verify against the repo before relying on details.