Repository navigation
docs(readme): search-friendly opening - #877
Conversation
The heading and the first paragraph now carry what people search for: Strix Halo, Ryzen AI MAX+ 395, Radeon 8060S (gfx1151), XDNA 2 NPU, HRX and HIP kernels, Lemonade Server, OpenAI-compatible API, model-agnostic, with "one binary to rule them all" up front. Two headings say what is under them. Every fact was already in this README; nothing else changed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
PR Reviewer Guide 🔍(Review updated until commit 441a391)Here are some key observations to aid the review process:
|
bong-water-water-bong
left a comment
There was a problem hiding this comment.
VERDICT: READY
At 7c6ec6062. README only, +8/-5, build/CodeQL/Analyze ×4/pr_agent all success at that head.
What I checked, beyond reading it:
The claims are already in the tree, so nothing new is asserted. Every term the new opening promotes is in docs/
already: "Strix Halo" 18 files, "Ryzen AI MAX+ 395" 3, "Radeon 8060S" 3, "gfx1151" 3, "XDNA 2" 4, "OpenAI-compatible" 5.
The four hunks touch the H1, the first paragraph, one ## heading and one ## heading — nothing else; the Milestones row
("v2026.44 is a draft") and the Apache header are untouched.
README.md is the source of 1bit.gg/overview.html, so I built the site against this README and against main's, which is
the consequence a README SEO change actually has:
ONEBIT_ENGINE_DIR=<engine at pr877> python3 tools/site.py <out> -> 53 pages, rc 0 (same as main's README)
render diff of overview.html: the H1, the first paragraph, and the two <h2>s — nothing else
<title> and <meta description> unchanged ("What the 1bit engine is and what runs where · 1bit engine")
one <h1> per page still holds (the site's own build check)
JSON-LD TechArticle.headline follows the H1 automatically -> "…local LLM inference on AMD Strix Halo (Ryzen AI MAX+ 395)"
Anchors. Nothing in this repo or in the site repo links to #what-it-is or #documentation (grep over *.md,
*.html, *.py, *.json), so the two headings can be renamed. Both anchors do change, though —
#what-it-is-one-model-agnostic-binary-for-the-gpu-and-the-npu, #documentation-and-measured-results — on GitHub and on
overview.html. That is the one cost this PR carries that it does not mention: any inbound link from outside the repo (a
Discord message, an already-indexed snippet, someone's bookmark) lands at the top of the page instead of the section. Not a
blocker, and unavoidable if the headings are to carry the terms, but worth one sentence in the body.
Merge order with #859: verified rather than assumed — git merge-tree pr877 pr859 returns rc 0, no conflict, which
matches the body's claim about the untouched lines between the two PRs.
Two nits, neither blocking:
- "One binary to rule them all" now opens the first paragraph and the "What it is" paragraph about ten lines below, so a
reader sees the same sentence twice in the top fold on GitHub and on the overview page. The first paragraph could keep
the search terms and lose the slogan, since the h2 below now repeats it. - The PR body ends with a "🤖 Generated with [Claude Code]" trailer. Public repo, so it is worth dropping from these.
|
Persistent review updated to latest commit 36d88f4 |
bong-water-water-bong
left a comment
There was a problem hiding this comment.
VERDICT: READY
At 36d88f4e4 (merge of eec6d2231 into 7c6ec6062), superseding my line at 7c6ec6062. README.md is
blob-identical at both heads, so the update added nothing and the review below it stands unchanged.
Evidence recap, since this one is about SEO and the evidence is not obvious from the diff: every promoted term already
appears in docs/ (Strix Halo 18 files, Ryzen AI MAX+ 395 3, Radeon 8060S 3, gfx1151 3, XDNA 2 4, OpenAI-compatible 5);
README.md is the source of 1bit.gg/overview.html and I built the site against this README (53 pages, rc 0, one <h1>,
<title>/description unchanged because they come from PAGE_META, JSON-LD headline following the new H1); nothing in the
repo or the site links to #what-it-is/#documentation; and merge-tree pr877 pr859 is clean.
CI at 36d88f4e4: build success (03:21Z), CodeQL/Analyze x4/pr_agent success.
|
Persistent review updated to latest commit 441a391 |
bong-water-water-bong
left a comment
There was a problem hiding this comment.
VERDICT: READY
At 441a391d4 (merge of aa6db8ded into 36d88f4e4), superseding my line at 36d88f4e4. README.md is
blob-identical (243fcbe24934… at both heads), so the merge added no author change and the review stands: the promoted
terms all already exist in docs/, README.md is the source of 1bit.gg/overview.html (built both ways — 53 pages rc 0, one
<h1>, <title>/description unchanged), nothing links to the two renamed anchors, and merge-tree pr877 pr859 is clean.
CI at 441a391d4: build success (03:30Z).
Part of the owner's SEO pass (site side: 1bit-MONSTER/site#18). README.md only; proposed, not for self-merge.
What changes
1bit engine: local LLMs on the AMD Ryzen AI MAX+ 3951bit engine: local LLM inference on AMD Strix Halo (Ryzen AI MAX+ 395)gfx1151) via HRX + our HIP kernels, XDNA 2 NPU via our NPU runtime, Lemonade Server + OpenAI-compatible API; ends with the same "Boot a USB stick ... That is the whole setup."## What it is## What it is: one model-agnostic binary for the GPU and the NPU## Documentation## Documentation and measured resultsGitHub and search engines use the H1 and the first paragraph as the repository's snippet, so those now carry the terms people search for (Strix Halo, Ryzen AI MAX+ 395, Radeon 8060S, gfx1151, XDNA 2 NPU, Lemonade Server, OpenAI-compatible, local LLM inference).
Kept true
tools/copyright.py --checkis unaffected.#what-it-isor#documentation.Merge order with #859
#859 rewrites the "Get started" steps (lines 30-33 plus an inserted paragraph). This PR touches lines 17, 25-26, 37 and 55, with unchanged lines between, so the two merge in either order without a conflict.
🤖 Generated with Claude Code