Reliability of post-edit LLMs on Pile-CC benchmark leaderboard · SOTA2 Research

Reliability of post-edit LLMs on Pile-CC

100BLEUGRACE

Evaluation Results

Method	Links
GRACE2025.06		100	100
GRACE2025.06		100	100
GRACE2025.06		100	100
GRACE2025.06		99.7	99.7
GRACE2025.06		99.7	99.7
GRACE2025.06		99.7	99.7
GRACE2025.06		99.7	99.7
GRACE2025.06		99.7	99.7
GRACE2025.06		99.7	99.7
PME2025.06		95.1	95.6
PME2025.06		94.6	95
PME2025.06		93.8	94.3
PME2025.06		92.7	93.6
MEMIT2025.06		91.2	91.4
PME2025.06		90.5	90.8
PME2025.06		89.8	90.7
PME2025.06		89.7	90.7
MEMIT2025.06		89.6	90.5
MEMIT2025.06		89.5	90.2
MEMIT2025.06		88.5	88.2
MEMIT2025.06		87.2	87.8
PME2025.06		87.2	87.9
PME2025.06		87	87.9
MEMIT2025.06		86.2	86.4
MEMIT2025.06		84.9	85.9
MEMIT2025.06		83.3	84
MEMIT2025.06		83.3	84
DeMem2025.06		82.8	84.6
DeMem2025.06		82	83.5
DeMem2025.06		81	82
DeMem2025.06		79.6	81.3
DeMem2025.06		78.8	79.7
DeMem2025.06		77.9	78.3
DeMem2025.06		72.6	73.2
DeMem2025.06		70.6	72.2
DeMem2025.06		69.4	71.2