[Docs] refactor Contribution Guide

[shuaills]

Motivation

related to #2662

[zhaochenyang20]

Great job!@

[zhaochenyang20]

 git clone https://github.com/<your_user_name>/sglang.git
 cd sglang

[zhaochenyang20]

Install Dependencies and Build SGLang on the Latest Commit

# Install flashinfer

pip install --upgrade pip
pip install -e "python[all]" --find-links https://flashinfer.ai/whl/cu124/torch2.4/flashinfer/

# Install latest SGLang

cd sglang/python
pip install .

Note: Please check the FlashInfer installation doc to install the proper version according to your PyTorch and CUDA versions.

[zhaochenyang20]

pip3 install pre-commit
cd sglang
pre-commit run --all-files

[zhaochenyang20]

Delete this line.

[zhaochenyang20]

pre-commit run --all-files will manually run all the configured checks on your repository, applying automatic fixes where possible. If it fails the first time, please run it again to make sure the lint errors are fixed. Ensure your code passes all these pre-commit link checks. Otherwise, there are grammar errors beyond style in your code. And do not commit to the main branch. Always checkout a new branch with a valid name and commit, then create a pull request on that branch.

[zhaochenyang20]

We prefer Jupyter Notebook over Markdown files since its correctness can be ensured by our docs CI.

First make a new file under approperiate path and modify index.rst if neccessary.

# Compile all .ipynb and markdown files
# The jupyter notebooks will be executed

make compile

# Generate static files and they can be found under _build/html
make html

# Preview the docs
# Open your browser at the specified port to view the docs locally

bash serve.sh

Once you are sure your docs are right and passed compile, make a PR.

# Clean Notebook outputs to make your PR clear

pip install nbstripout
find . -name '*.ipynb' -exec nbstripout {} \;

# Run pre-commit

pre-commit run --all-files

Then you can submit your PR.

TODO: How to run server/engine and close server/engine in docs.

[zhaochenyang20]

delete.

[zhaochenyang20]

• Create new test files under /test, and ensure they are referenced in /test/srt/run_suite.py or /test/lang/run_suite.py, other they will not be picked up in CI.

[zhaochenyang20]

• In CI, all tests are executed automatically. You may modify the corresponding workflow in /.github/workflows if you wish to include custom test groups or additional checks.

[zhaochenyang20]

Introduce this:

https://github.com/zhaochenyang20/Awesome-ML-SYS-Tutorial/tree/main/sglang/code-walk-through

[zhaochenyang20]

Also modify https://sgl-project.github.io/backend/openai_api_completions.html

Great job!!!

[zhaochenyang20]

delete >

[zhaochenyang20]

delete our

[zhaochenyang20]

Install correct version of flashinfer according to your local xxx

[zhaochenyang20]

Below is an example on PyTorch 2.4 with CUDA 12.4:

[zhaochenyang20]

delete

Example: Installing FlashInfer (Pick the right version for your CUDA/PyTorch)

[zhaochenyang20]

delete this

[zhaochenyang20]

Install the lateset SGLang on your own fork repo

[zhaochenyang20]

Make them into one line.

[zhaochenyang20]

If you need to run and shut up a SGLang server or engine, following these examples:

1. Launch and close Sever:

#Launch Sever

from sglang.utils import (
    execute_shell_command,
    wait_for_server,
    terminate_process,
    print_highlight,
)

server_process = execute_shell_command(
    "python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3.1-8B-Instruct --port 30000 --host 0.0.0.0"
)

wait_for_server("http://localhost:30000")

# Terminate Sever

terminate_process(server_process)

2. Launch Engine

# Launch Engine

import sglang as sgl
import asyncio

llm = sgl.Engine(model_path="meta-llama/Meta-Llama-3.1-8B-Instruct")

# Terminalte Engine
llm.shutdown()