Lightweight LLMs for 3GPP Specifications: Fine-Tuning, Retrieval-Augmented Generation and Quantization

Autor(es)

de Arimatea, Jose ; Lopes Junior, Passos ; Pereira, Jayr ; do Carmo, Diedre Santos ; Lotufo, Roberto ; Rothenberg, Christian Esteve

• Tipo de documento: Artigo Científico
• Fonte: 2025 IEEE 11TH INTERNATIONAL CONFERENCE ON NETWORK SOFTWARIZATION, NETSOFT; v. N/A, p. 6-pg., 2025-01-01.

Resumo

Interpreting complex 3GPP telecommunications standards for question and answering (QA) poses a challenge for general-purpose LLMs due to their specialized terminology and high computational demands, limiting their use in resource-constrained environments. This work explores an efficient, open-source approach using the TeleQnA dataset of 10,000 telecom questions and the TSpec-LLM repository of processed 3GPP documents. We enhance a lightweight Llama 3.2 (3B parameters) model, quantized from 16-bit precision to 4 bits, through fine-tuning and RAG to improve accuracy without heavy resource reliance. Unlike prior resource-intensive or proprietary solutions, our method reduces memory demands, enabling deployment on modest hardware like edge devices or softwarized networks. Shared via GitHub repositories [1], this approach advances cost-effective, reproducible AI for telecommunications QA, supporting contexts where budgets, computation, or public internet access are limited. (AU)

• Processo FAPESP: 21/00199-8 - Redes e serviços inteligentes rumo 2030 (SMARTNESS)
• Beneficiário: Christian Rodolfo Esteve Rothenberg
• Modalidade de apoio: Auxílio à Pesquisa - Programa Centros de Pesquisa Aplicada