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

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.