Thank you for your nice job! I would very appreciate it if you could take another new editing method into consideration.
As mentioned in "[IBM] Larimar: Large Language Models with Episodic Memory Control", GPM(Generative Pseudo Memory) mechanism is used for editing LLM, whose wall clock time for each editing method across 10 edits is shorter than ROME and GRACE etc.
Additionally, Larima's capacities of single fact editing, sequential fact editing, selective forgetting and generalization to long input context are comparable or even better than competitive baselines.
Hi,
Thank you for your nice job! I would very appreciate it if you could take another new editing method into consideration.
As mentioned in "[IBM] Larimar: Large Language Models with Episodic Memory Control", GPM(Generative Pseudo Memory) mechanism is used for editing LLM, whose wall clock time for each editing method across 10 edits is shorter than ROME and GRACE etc.
Additionally, Larima's capacities of single fact editing, sequential fact editing, selective forgetting and generalization to long input context are comparable or even better than competitive baselines.
Looking for your reply!