mello-requires-no-weight-access

IN premise — summaries/2026/08/24/zhong-2023-mquake-s6-related-work.md

Created 2026-08-25T02:59:06+00:00

MeLLo requires no gradient updates, no fine-tuning, and no access to model weights; it operates purely via prompting and retrieval, making it applicable to black-box LMs like GPT-3.5.

Summary

MeLLo works entirely through prompting and information lookup, never touching a model's internal parameters or requiring any retraining. This means it can be dropped onto any language model accessible only through an API, including closed-source ones, without needing the vendor to expose its architecture or allow weight modifications.