Skip to content

feat(news): split news from blog, credit authors with GitHub avatars, import KDA² - #6

Merged
futrime merged 5 commits into
mainfrom
claude/blog-news-split
Oct 5, 2026
Merged

futrime merged 5 commits into
mainfrom
claude/blog-news-split

Conversation

@futrime

@futrime futrime commented Oct 5, 2026 •

Copy link
Copy Markdown
Member

Splits result reports out of the blog into a new News section, imports KDA² as the first Blog post, and puts the GitHub avatars of the people who did the work on every byline.

What changes

  • News / Blog split. The 16 result posts move from blog/ to news/ with the same slugs. Six essays stay in the Blog: launch, four-layers, review-is-next-prompt, model-tool-flow, sglang-omni-numerics and sglang-diffusion-quality-tiers. Each section gets its own index, nav item, sidebar and RSS feed (/blog/feed.rss, /news/feed.rss). Both feeds get rel=alternate and a footer link. feed.mjs checks each feed and catches a post filed in the wrong one.
  • Redirects. public/blog/<moved-slug>.html stubs send the old URLs on to /news/<slug> and keep the #hash. public/results/* now points at /news/. Cross-links in projects/* and HomeLanding.vue are retargeted, and the home reel shows news only.
  • Authors with faces. theme/people.ts maps each name to a GitHub numeric id and handle. PostMeta and PostMosaic render the avatars, and fall back to initials when there is no verified account.
  • KDA² import. blog/2026-09-27-kda-for-kda.md credits all 13 authors. It sets the canonical link to nvlabs.github.io in both the frontmatter and <head>. The eight charts are stills captured from the original page at 1060px (2x), with alt text taken from the original aria-labels. Tapping a chart opens it at full size. Links to humanfia/humanize2 now point to humanfia/humanize.
  • Fixes from code review. The about page's "following along" line and the team page's Credit line now point at News as well as the Blog.

News titles, old → new

Old New
One person, some agents, and gem5's build system 567 files, one engineer: agents move gem5 from SCons to CMake
Top three on every track at the MLSys FlashInfer contest 33× on DSA: KDA kernels place top three on every MLSys 2026 track
Forty operators into SGLang SGLang merges 40+ agent-tuned operators, one lifting throughput 71%
KDA generalizes to hardware it was never tuned for Thin docs, no corpus: KDA still finds speedups on AMD hardware
Fifty-three firsts on SOL Bench, from one node One node, one week, nobody watching: 53 first places on SOL Bench
670 of 672 on PutnamBench 670 of 672 on PutnamBench: a Codex loop passes two purpose-built provers
First on SOLExec Bench, L1 single operations First on SOLExec L1 — by 0.0024, on a track everyone had tuned
Six of six at IMO 2026, on two different backends 6 of 6 at IMO 2026, Lean-checked and 3.2× faster than AxiomProver
Lean-Eval: 149 of 219, and second place #2 on Lean-Eval: a general loop finishes 6 problems behind a prover
Physics and quantum, formalized end to end 23/23 IPhO 2026 subproblems and 36/36 quantum tasks, proved in Lean
QCode Discovery, made fail-closed No certificate, no win: QCode Discovery's search now fails closed
KDA 1.5 goes past human SOTA on every contest track KDA 1.5 beats the best human kernels on every MLSys track, up to 1.39×
ProgramBench: 3.5%, against 0.5% and 0% 0% and 0.5% alone, 3.5% together: a review loop on ProgramBench
The MSA indexer, 3.3× faster in production MSA indexer: 6.5× faster prefill in production, bitwise identical
Ten workflows, nineteen competitions 14 of 19 Kaggle competitions in the top 5%, five inside the top 1%
Lean-Eval: first place, and 172 research-level proofs From #2 to #1 on Lean-Eval in three weeks: 172 research-level proofs

Author evidence

  • Kaggle post: added Menghan Li (44 commits) and Yitong Liu (14 commits) to agentkaggle/submissions.
  • Jin Pan = jhinpan: author of FlyDSL #711 and #685, which the ROCm post cites.
  • Lesheng Jin: all 27 commits in humanfia/sol-execbench-kda.
  • gem5: build: replace SCons with CMake+Ninja and Bazel build systems gem5/gem5#2969 is by SihaoLiu. The post now links to it.
  • SGLang: #22664, #27431, #29708, #29390, #29361, #33453, #28166 and #28708 are by BBuf (Xiaoyu Zhang). The SGLang post keeps BBuf as its only byline because it summarises his article, even though 4 of its 6 PRs are by others.
  • Zhengyang Zhang: IMO 2026 and Lean-Eval. Hongzhou Lin: lean-eval-code.
  • KDA² authors: the names come from the original post. Hongyi Jin also committed to humanfia/kda-for-kda-release. Zihao Ye, Junru Shao and the remaining authors are TVM, MLC and FlashInfer maintainers whose profile names match.
  • Not added or not linked:
    • Ligeng Zhu is not added to older posts, because his commits come after them.
    • Yuchen Jin has no GitHub account tied to the work, so he shows as initials.
    • Avery Huang = averyhNV is a medium-confidence match.
  • PutnamBench June post: gets a "since moved" note (672/672, joint first).

Known trade-off

/blog/feed.rss used to carry results. From now on it carries essays only, and results go to /news/feed.rss under new item links. Feed readers don't follow HTML redirects, so people subscribed to the old feed have to subscribe to the news feed to keep getting results. The blog index, the footer and the about page all link it.

Verification

  • pnpm build, check:anchors, check:feed and check:docs are all green.
  • Ran pnpm preview with Playwright. Took screenshots of /blog/, /news/, KDA² and two news posts in light and dark, at 1440px and 375px.
    • No images are broken.
    • Every avatar loads (13 on KDA², 21 on /news/).
    • There is no horizontal scroll.
  • All 32 old URLs (/blog/<slug> and /blog/<slug>.html for each of the 16 moved posts) land on /news/<slug> with the hash kept.
  • /results/* lands on the matching news post.

🤖 Generated with Claude Code

@futrime
futrime marked this pull request as ready for review October 5, 2026 10:02
Copilot AI balanced review requested due to automatic review settings October 5, 2026 10:02

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

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

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

futrime and others added 5 commits October 5, 2026 19:19
WIP: KDA² import and verification still to come.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@futrime
futrime force-pushed the claude/blog-news-split branch from 31227cb to 1e861b7 Compare October 5, 2026 11:21
@futrime
futrime requested a review from Lyken17 as a code owner October 5, 2026 11:21
@futrime
futrime merged commit b6961cf into main Oct 5, 2026
7 checks passed
@futrime
futrime deleted the claude/blog-news-split branch October 5, 2026 11:22
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.

2 participants