Verfahren und System zur Optimierung der Feinabstimmung eines Grosssprachenmodells (llm)

Details

Veröffentlichungs-Nr.

EP4756678

Anmeldetag

4. Dezember 2025

Veröffentlichung

10. Juni 2026

Rechtsraum

EP

Abstract

Method, system, and computer-readable storage media for optimizing fine-tuning of a Large Language Model (LLM) are disclosed. An adapter recommendation is generated for fine-tuning of the LLM based on input data and a desired output criterion corresponding to each of output aspects. Based on the adapter recommendation, initial coefficient values for coefficients that correspond with the output aspects and an initial ranking value are assigned. Once the coefficients and ranking value are assigned, an adaptive multi-objective low rank adaptation and uncertainty estimation are iteratively performed for a predefined number of iterations or until the desired criterion corresponding to each of the output aspects is achieved. Upon reaching the predefined number of iterations or achieving the desired criterion corresponding to each of the output aspects, an output value for each of the output aspects is derived. The derived output value is used for fine-tuning of the LLM.

Anmelder

Firma

Accenture Global Solutions Limited

Land

🇮🇪 Irland