Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

Mass-Editing Memory in a Transformer

About

Recent work has shown exciting promise in updating large language models with new memories, so as to replace obsolete information or add specialized knowledge. However, this line of work is predominantly limited to updating single associations. We develop MEMIT, a method for directly updating a language model with many memories, demonstrating experimentally that it can scale up to thousands of associations for GPT-J (6B) and GPT-NeoX (20B), exceeding prior work by orders of magnitude. Our code and data are at https://memit.baulab.info.

Kevin Meng, Arnab Sen Sharma, Alex Andonian, Yonatan Belinkov, David Bau• 2022

Related benchmarks

TaskDatasetResultRank
Multitask Language UnderstandingMMLU (test)
Accuracy21.83
303
Lifelong Free-text Knowledge EditingMRLF-Bench
BLEU36.36
140
Knowledge EditingzsRE
Generality96.4
110
Knowledge EditingCounterFact
Efficacy9.38e+3
91
Commonsense Question AnsweringCommonsenseQA
Accuracy20.23
81
Privacy EditingTDE Email
Leakage0.00e+0
56
Privacy EditingTDE URL
Leakage0.00e+0
50
Training Data Extractionemail PII
Leakage0.00e+0
45
Training Data Extractionphone PII
Leak Count1
45
Training Data ExtractionURL PII
Leakage11
45
Showing 10 of 95 rows
...

Other info

Follow for update