ResearchBenchmarksEdit Reversal on Edit Reversal (test)Follow82RSRROME67.4471.227578.78Apr 18, 2026Evaluation ResultsMethodMethodLinksRSRTop-1 AccuracyROMEModel=GPT-2 XL, Pruned...Model=GPT-2 XL, Pruned (%)=10.0%2026.048277ROMEModel=LLaMA-3.2, Prune...Model=LLaMA-3.2, Pruned (%)=10.0%2026.047972MEMITModel=LLaMA-3.2, Prune...Model=LLaMA-3.2, Pruned (%)=8.8%2026.047865MEMITModel=GPT-2 XL, Pruned...Model=GPT-2 XL, Pruned (%)=4.5%2026.047478MEMITModel=Qwen-2.5, Pruned...Model=Qwen-2.5, Pruned (%)=7.9%2026.047060ROMEModel=Qwen-2.5, Pruned...Model=Qwen-2.5, Pruned (%)=15.3%2026.046866