This repository is a collection of documentation for Ascend supported open source projects.

• index.rst: the index of the opensource project doc.
• quick_start.rst: 5 minutes or 600 words, quickstarts enable people to quickly complete a discrete, focused task by illustrating a workflow with only essential steps. See also in Github quickstart guide.
• install.rst: installation guide, including common installation (such as binary, pypi, conda) for developers and a quick installation verification
• tutorial.rst: more complex than quickstart, help people learn about products and solve real world problems by guiding them through the entire workflow to complete a task. See also in Github tutorial guide.
• faq.rst: A frequently asked questions (FAQ) list.

Some communities already maintain Ascend/NPU documentation on their own sites. Those projects are linked from the homepage cards in index.rst instead of being built locally. Examples: verl, LLaMA-Factory, ms-swift, VeOmni, vllm-ascend, Triton-Ascend, DeepSpeed, ONNX Runtime, SGLang, LMDeploy, ROLL, and Twinkle.

To add such a project, create a lightweight sources/<project>/index.rst with links to the official docs, and add it to the hidden toctree in index.rst so it appears in the sidebar.

Before make html, the build fetches ascend_config.json via scripts/fetch_ascend_data.py (see fetch-config in ./Makefile). No git submodules or document copying is involved.

If you are interested in the project, please feel free to contribute it.

We use sphinx and rst file to create docs and you need to refer to rst syntax for help. Here is how you can create a new docs:

1. Create a rst file. You can put it anywhere, but it must be better to classify it.
2. Add the rst into the index.rst

When the development is ready, you can check and test it locally by following the directives below:

1. Install the required dependencies:

pip install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple --extra-index-url https://download.pytorch.org/whl/cpu

2. Building:

make html

3. Server on localhost:

python -m http.server -d _build/html 4000

Then open localhost:4000 in your browser.