How this mdBook is organized — for editors adding or moving pages.
src/
intro.md # Landing page (WIP banner)
SUMMARY.md # Sidebar navigation — edit this when adding pages
setup/ # Hardware and ROCm install
hardware.md
getting-started.md
installing-rocm.md
hsa-override.md
tuning/ # V620-specific host tuning
power.md
ecc.md # Optional Pro-card ECC disable (~2 GB)
p2p.md
vllm/ # vLLM serving on RDNA
overview.md # Section hub — start here
running.md # Images, docker run, quick start
recipes.md # Which stack / model / card count
configuration.md # Env vars, CUDA graphs, compose
quantization.md # GPTQ/AWQ, KV, MTP, INT4
fork.md # rdna_extras kernels
images.md # Building Docker images
llama-cpp/ # llama.cpp and RDNA2 fork
overview.md # Section hub
building.md # Stock llama.cpp build
rdna2-overview.md # Fork intro, build, requirements
rdna2-benchmarks.md # Author-reported numbers
rdna2-speculative.md # DFlash2, MTP, ngram
rdna2-serving.md # Launch, Docker, limits
reference/ # Lookup tables
env-vars.md # Environment variable cheat-sheet
verification.md # What's verified vs community-reported
troubleshooting/ # Problem → fix, by stack
index.md # Hub
general.md
vllm.md
llama-cpp.md
meta/ # Wiki maintenance
resources.md # External links
contributing.md
structure.md # This page
| Rule | Example |
|---|
| One topic per file — split when a page exceeds ~200 lines or mixes concerns | vllm/configuration.md vs vllm/quantization.md |
Section overview.md — first page in each major folder links to sub-pages | vllm/overview.md, llama-cpp/overview.md |
| kebab-case filenames | getting-started.md, not getting_started.md |
Relative links — use ../ to cross folders | From vllm/running.md to setup: ../setup/hsa-override.md |
| WIP banner — on pages with Discord/community claims | See vllm/overview.md |
Verification row — add/update in reference/verification.md for new benchmarks | Status: solid / fork-source / community / needs verify |
- Pick the folder (or create one if it’s a new major topic).
- Create
src/<folder>/<topic>.md with a # title and optional WIP banner.
- Add an entry to
src/SUMMARY.md under the right section.
- Link from the section
overview.md if one exists.
- Add verification rows if the page contains benchmarks or unconfirmed claims.
- Run
mdbook build — fix broken links before merging.
- Move shared intro to
overview.md or trim the original.
- Create focused sub-pages; link between them at the bottom (“Next steps”).
- Update
SUMMARY.md to nest sub-pages under the section.
- Grep for old filename links:
rg 'old-name\.md' src/
- Update
reference/verification.md section headers to match new paths.
| Content type | Location |
|---|
| Install ROCm, hardware list | setup/ |
| Power cap, P2P enablement, Pro ECC | tuning/ |
docker run, compose, env vars for vLLM | vllm/ |
| Which image / model / card-count recipe | vllm/recipes.md |
| Kernel source, fork branches | vllm/fork.md |
| Docker image build (vllm-rdna-docker) | vllm/images.md |
| Stock llama.cpp cmake build | llama-cpp/building.md |
| Fork benchmarks, DFlash2, TP serve | llama-cpp/rdna2-*.md |
| Env var tables | reference/env-vars.md |
| Claim audit / WIP status | reference/verification.md |
| Error messages and fixes | troubleshooting/ |
| External links (official / Docker / upstream / community) | meta/resources.md |