Skip to content

docs: add a quickstart from install to first chat reply - #1071

Merged
pmoutsias-amd merged 3 commits into
release/v0.3.0from
docs/0.3.0-combined
Nov 16, 2026
Merged

pmoutsias-amd merged 3 commits into
release/v0.3.0from
docs/0.3.0-combined

Conversation

@pmoutsias-amd

@pmoutsias-amd pmoutsias-amd commented Nov 12, 2026 •

Copy link
Copy Markdown
Contributor

Adds a Quickstart that takes a new user from installing ROCm CLI to a chat reply from a locally served model.

What changes

  • Adds a Quickstart section to the README and to the top of the Getting started page on the docs site: install, rocm examine, rocm install sdk, rocm serve, rocm chat, then stop the server. The site includes the README text, so there is one source.
  • Splits the serve and chat commands by engine. The qwen alias resolves to a GGUF model that only Lemonade serves (Ryzen AI, Radeon). Instinct GPUs serve through vLLM, so the Quickstart uses Qwen/Qwen2.5-1.5B-Instruct there. Readers pick their path from the effective_default_engine line in rocm examine. vLLM isn't available on native Windows, so Windows always uses Lemonade.
  • rocm chat --provider local without --model only matches the built-in qwen service, so the vLLM path passes --model.
  • Corrects the Configure ROCm and serve a model section, which said qwen serves out of the box on any GPU.
  • Moves the CLI demo recording from the First run section to the Quickstart, and updates the Getting started card on the site index to mention the Quickstart.

Testing

  • The Sphinx docs build (-W), lychee link check and Read the Docs preview pass in CI.
  • The commands, flags, rocm examine field names and the endpoint:/logs:/stop: lines in the deployment summary are checked against the CLI source.
  • Docs only. No CLI behavior changes, so no new e2e scenario.

@pmoutsias-amd

Copy link
Copy Markdown
Contributor Author

Update: quickstart follow-ups (6860a13)

  • The landing card now names the quickstart.
  • Step 2 shows the two rocm examine lines to look for, including the vllm value on Instinct GPUs.
  • The intro states the supported platforms (Linux, WSL2, Windows) and warns about the first-run downloads, including the roughly 2.5 GB qwen model.
  • Step 6 shows the form of the stop command: rocm services stop <service-id> --yes.
  • The CLI demo now sits directly after the quickstart.

Docs-only change, so no scenario applies. Sphinx -W builds cleanly. The quickstart has not been run end to end on a GPU.

Add a Quickstart section to the README and the top of the Getting started
page. It walks through rocm examine, rocm install sdk, rocm serve, and
rocm chat, with a separate serve and chat command for each engine: the
qwen alias is a GGUF model that only Lemonade serves, so Instinct GPUs
on vLLM need a safetensors model.

Also correct the Configure ROCm section, which said qwen serves out of
the box on any GPU.

Signed-off-by: pmoutsia_amdeng <peter.moutsias@amd.com>
Name the quickstart on the landing card, show the two `rocm examine`
lines the reader should look for, state the supported platforms and the
size of the first download up front, show the form of the stop command,
and put the CLI demo right after the quickstart.

Signed-off-by: pmoutsia_amdeng <peter.moutsias@amd.com>
@pmoutsias-amd pmoutsias-amd changed the title docs: add a quickstart and combine the 0.3.0 docs updates docs: add a quickstart from install to first chat reply Nov 16, 2026
@pmoutsias-amd
pmoutsias-amd marked this pull request as ready for review November 16, 2026 03:19
@pmoutsias-amd
pmoutsias-amd requested a review from a team as a code owner November 16, 2026 03:19

@juhovainio juhovainio left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks good. I checked the quickstart's commands, flags and field names against the code (examine output, chat flags, --model matching, rocm services stop <id> --yes, the qwen alias) and they match; links follow the repo's relative-link rules and CI is green.

One small thing, left inline: the "about 2.5 GB" download size isn't backed by anything in the repo. Also worth a quick grep of docs/*.md and docs/manual-testing.md for leftover "qwen serves out of the box" wording. This reflects a read of the diff and the code, not a live run on a GPU.

Comment thread README.md Resolved
Comment thread docs/rocm-docs/sphinx/_toc.yml.in Resolved
Comment thread docs/rocm-docs/install/installation.md Resolved
Comment thread docs/rocm-docs/install/installation.md Resolved
Comment thread docs/rocm-docs/install/installation.md Resolved
@@ -6,15 +6,43 @@ SPDX-License-Identifier: MIT

# Getting started with ROCm CLI

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

As suggested in the comment on TOC, two headings are not required here. We can call this doc "Quickstart guide" or something similar.

Comment thread README.md Resolved
Comment thread README.md Resolved
Comment thread README.md Resolved
Comment thread README.md Resolved
Comment thread docs/rocm-docs/index.md Resolved
Say "a few GB" for the first model download instead of a figure that
isn't sourced in the repo, and cover both engines since the vLLM model
is a different size. Reword the intro and add a lead-in before the
numbered steps.

Signed-off-by: pmoutsia_amdeng <peter.moutsias@amd.com>
@pmoutsias-amd
pmoutsias-amd requested a review from SwRaw November 16, 2026 15:20
@pmoutsias-amd
pmoutsias-amd merged commit d67eb8d into release/v0.3.0 Nov 16, 2026
32 checks passed
@pmoutsias-amd
pmoutsias-amd deleted the docs/0.3.0-combined branch November 16, 2026 15:20
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants