memit-mend-vs-memit-runtime-10k

IN premise — summaries/2026/08/24/meng-2022-memit-sR-references-chunk-1.md

Created 2026-08-25T02:58:13+00:00

For 10,000 edits on GPT-J, MEND takes 98.25 seconds, ROME takes ~12.29 hours (~4 s/edit sequential), and MEMIT's z_i computation takes ~6.54 hours (serial) with parallel insertion taking ~0.90 hours.

Summary

When you need to make ten thousand edits to a GPT-J model, the choice of editing method changes total wall-clock time from under two minutes (MEND) to over twelve hours (ROME), which is roughly a 450x difference. This means that for large-scale model correction, runtime is the dominant practical constraint: ROME is essentially infeasible at this scale, MEMIT offers a middle ground by parallelizing part of the work but still has a long serial bottleneck, and MEND is the only option that completes in minutes.