Traces v1.1: TiKV reruns and v1.1 label scores
#1
by Essoz - opened
Mirrors OrderLab/distdebug-bench#4. tikv-17830 and tikv-19220 are rerun on their corrected snapshots for 14 configurations (the o4-mini rows keep their original attempts); every score uses the v1.1 labels. configs.json lists Kimi K2.6 above MiMo v2.5 Pro to match Table 1, and harness versions are recorded only where the run recorded them. paper/reproduce.py on this revision reproduces the v1.1 reference exactly.
Essoz changed pull request status to open
Essoz changed pull request status to merged