Conversation
|
Final verification passed on the actual Read the Docs deployment: 128 live checks for EN/ZH, desktop/mobile, light/dark, updated Long/Overall scores, seed disclosure, CSV, all 11 cost rows, resource isolation and native fallback. All assets resolve to misc@705bd44; no local resource substitution. Python 3.10/3.11/3.12 unit tests, pre-commit and both documentation builds are green. Preview: Chinese Performance, Chinese costs, English Performance. Please merge misc#69 first. |
wilburx813
left a comment
wilburx813
left a comment
There was a problem hiding this comment.
Add markers next to models to link leaderboard entries with their evaluation notes？it's hard to associate leaderboard with note now.
qurakchin
left a comment
qurakchin
left a comment
There was a problem hiding this comment.
- please clean the target50_codex_results.md file.
- cc reviewed and says"scope-note 在离线 fallback 中是空类（docs.css 未定义）；benchmark-method 规则成为死代码；pin 的 misc@705bd44 是悬空 commit，misc#69 仍未合并，若以 squash/rebase 合并该 SHA 会被 GC，届时全部资产 URL 失效。"
869a96c to
d30b346
Compare
|
Following up on this review: Removed the old result table and its table-parsing test in d30b346. Both guides keep a link to the archived report; the v1 protocol checks remain. The missing scope-note style and unused benchmark-method rule are fixed in misc#69 (bb484a5). The asset pin is still 705bd44 for now: after #69 is merged, I will pin the actual misc main commit and verify the public CDN before this PR is merged. |
|
Following up on this review: Added matching numbered links beside the relevant model names in the charts, full tables and English/Chinese fallback. They lead to the existing starred notes and work with keyboard navigation. The renderer change is in misc#69 (bb484a5), with the documentation fallback in d30b346. Scores and CSV contents are unchanged. |
3 participants
The leaderboard, tutorials and README comparisons now use the same published scores and evaluation notes. The current RoboCasa GPT-5.5 result is 57.1%; the earlier per-task report is linked at its historical commit instead of duplicated in this branch.
Models with evaluation notes have numbered links beside their names in the charts, full tables and English/Chinese static fallback. The existing starred notes remain intact. The obsolete result table and its table-parsing test are removed; the v1 protocol and its checks remain.
Validation: 590 unit tests passed, 3 skipped; pre-commit and strict English/Chinese Sphinx builds passed. Local browser checks passed across desktop/mobile, light/dark, both languages, offline and no-JavaScript fallback. All 28 compared CSV outputs are byte-identical before and after the note-link changes.
Dependency: RLinf/misc#69 supplies the renderer and fallback styles. This branch still pins
705bd44; the updated assets were tested by serving the candidate files locally. After #69 is merged by a maintainer, the URLs must be updated to its actualmaincommit and checked against the public CDN before this PR is merged.