| Name | Modified | Size | Downloads / Week |
|---|---|---|---|
| Parent folder | |||
| README.md | 2026-10-19 | 881 Bytes | |
| v0.5.21 source code.tar.gz | 2026-10-19 | 1.1 MB | |
| v0.5.21 source code.zip | 2026-10-19 | 1.2 MB | |
| Totals: 3 Items | 2.3 MB | 0 | |
Fix Artificial Analysis checkpoint matching for Kimi K2 and Command R+ (#188).
The Kimi K2 row now maps only to moonshotai/Kimi-K2-Instruct. The April Command R+ row maps only to CohereForAI/c4ai-command-r-plus. Kimi-K2-Base and c4ai-command-r-plus-08-2024 no longer inherit their sibling's live AA score. Stored fallback scores are unchanged.
Use --refresh to bypass an existing benchmark cache.
The release candidate passed all 594 tests on Python 3.11, 3.12, and 3.13, plus Ruff lint and format checks. Its wheel was built from the sdist and checked against all 88 source package files. A clean Python 3.13 installation with production dependencies only returned recommendation JSON from an empty cache and the same picks on a cached rerun.
Required lint and Python 3.11, 3.12, and 3.13 CI checks pass on release commit 9f3709fbbe468c6ab2dd676d6a3e9888cff62562.
