[Feature] docs: Improve documentation on how to use EAGLE speculative docoding

[daviddl9]

Checklist

• 1. If the issue you raised is not a feature but a question, please raise a discussion at https://github.com/sgl-project/sglang/discussions/new/choose Otherwise, it will be closed.
• 2. Please use English, otherwise it will be closed.

Motivation

The recent addition of EAGLE speculative decoding in here is powerful. Thank you for creating and maintaining such a useful tool! The existing codebase gives insufficient examples of how it can be used (e.g for Llama3 models, for example) together with docker compose. It would be great if another file like https://github.com/sgl-project/sglang/blob/main/docker/compose.yaml can be added to illustrate how the feature can be used in docker environments. Thanks for looking into this issue!

[zhaochenyang20]

Yeah. First we should have a docs on EAGLE.

[simveit]

I can take this @zhaochenyang20

[zhaochenyang20]

@simveit Hey Simon. Thanks so much. We can work on accuracy tests first. I will let Jin @jhinpan to take this~

[zhaochenyang20]

1. Add docs for eagle first. Append one ipynb in docs/backend/speculative_decoding.ipynb following this docs: update SGLang status SafeAILab/EAGLE#173 and the style should follow https://docs.sglang.ai/backend/structured_outputs.html
2. Later check what we need to do for the docker compile.

[zhaochenyang20]

Detailed guidance are here:

https://docs.sglang.ai/references/contribution_guide.html#writing-documentation-running-docs-ci

and set up your enviroments.

[b8zhong]

Is in https://docs.sglang.ai/backend/speculative_decoding.html 👍