The paper "TelecomGPT: A Framework to Build Telecom-Specific Large Language Models" addresses the challenge of adapting general-purpose Large Language Models (LLMs) to the telecommunications domain. The authors propose a pipeline that includes continual pre-training, instruction tuning, and alignment tuning to create a telecom-specific LLM, named TelecomGPT. They collect and build telecom-specific datasets for pre-training, instruction tuning, and alignment tuning, and extend existing evaluation benchmarks to include new telecom-specific tasks such as Telecom Math Modeling, Telecom Open QnA, and Telecom Code Tasks. The results show that TelecomGPT outperforms state-of-the-art (SOTA) LLMs like GPT-4, Llama-3, and Mistral in Telecom Math Modeling and achieves comparable performance in other benchmarks. The paper also discusses the challenges of building telecom-specific LLMs, including the lack of relevant datasets, the dominance of technical documents, multi-modality content, and frequent knowledge updates. The contributions of the paper include the development of a telecom-specific LLM, the creation of new evaluation benchmarks, and the proposal of a training pipeline to adapt general-purpose LLMs to the telecom domain.The paper "TelecomGPT: A Framework to Build Telecom-Specific Large Language Models" addresses the challenge of adapting general-purpose Large Language Models (LLMs) to the telecommunications domain. The authors propose a pipeline that includes continual pre-training, instruction tuning, and alignment tuning to create a telecom-specific LLM, named TelecomGPT. They collect and build telecom-specific datasets for pre-training, instruction tuning, and alignment tuning, and extend existing evaluation benchmarks to include new telecom-specific tasks such as Telecom Math Modeling, Telecom Open QnA, and Telecom Code Tasks. The results show that TelecomGPT outperforms state-of-the-art (SOTA) LLMs like GPT-4, Llama-3, and Mistral in Telecom Math Modeling and achieves comparable performance in other benchmarks. The paper also discusses the challenges of building telecom-specific LLMs, including the lack of relevant datasets, the dominance of technical documents, multi-modality content, and frequent knowledge updates. The contributions of the paper include the development of a telecom-specific LLM, the creation of new evaluation benchmarks, and the proposal of a training pipeline to adapt general-purpose LLMs to the telecom domain.