Fix #2090: fix: README benchmark scores in News section don't match Performance table (LoCo#2092
Conversation
…mance table The News section for 2026-07-02 quoted LoCoMo=92.34 and LongMemEval=93.40 (added in MemTensor#2019), while the Performance table just below reports 88.83 and 89.20 (added in MemTensor#2078 with the 10-benchmark 2026-07-02 evaluation report). Update the News entry in both README.md and README_ZH.md to 88.83 / 89.20 so the two sections agree; the Performance table remains the authoritative source. Closes MemTensor#2090
🤖 Open Code ReviewTarget: PR #2092 ✅ OpenCodeReview: No supported files changed. Generated by cloud-assistant via Open Code Review. |
shinetata
left a comment
There was a problem hiding this comment.
Independently verified: this is the correct fix for #2090. The News item (added in #2019, dated 2026-07-02) and the Performance table (added in #2078, same date, "10 benchmark scores from the 2026-07-02 evaluation report") reported different LoCoMo/LongMemEval numbers for the same benchmarks. This diff aligns the News item with the Performance table (88.83 / 89.20), which is the more completely-sourced number (10 benchmarks, linked to OmniMemEval methodology).
Diff is minimal and scoped to the two affected lines in README.md / README_ZH.md. CI is green across the full matrix. Approving.
Description
Fixed the README score mismatch (LoCoMo 92.34→88.83, LongMemEval 93.40→89.20) in both
README.mdandREADME_ZH.md, aligning the News section with the Performance table. Committed on the working branch (64cddbfb) and pushed toorigin. Task spec archived to the siblingmemos-autodev-specsrepo (main@d8af691).Now submitting the completion envelope so the scheduler can open the PR.
Sources:
Related Issue (Required): Fixes #2090
Type of change
Please delete options that are not relevant.
How Has This Been Tested?
Automated tests are pending.
Checklist
@MatthewZhuang, @CarltonXiang, @syzsunshine219, @World-controller please review this PR.
Reviewer Checklist