The increasing interest in Large Language Models (LLMs) in the telecommunications sector highlights their potential to revolutionize operational efficiency. However, their deployment is hindered by their large size and high computational demands, making them unsuitable for resource-constrained environments. Recent advancements have led to the development of small language models (SLMs) that perform comparably to larger models in tasks like coding and common-sense reasoning. Phi-2, a compact yet powerful SLM, is an example of this trend. This paper evaluates Phi-2's understanding of the telecommunications domain and enhances its capabilities using a Retrieval-Augmented Generation (RAG) approach, integrating an extensive knowledge base of telecom standards. The enhanced Phi-2 model demonstrates improved accuracy in answering telecom standard questions, rivaling the performance of GPT-3.5. The paper also explores Phi-2's capabilities in problem-solving scenarios within the telecom sector, highlighting its potential and limitations. The study compares Phi-2 with larger models like GPT-3.5 and GPT-4, using the TeleQnA dataset to assess their telecom knowledge. Phi-2 achieves an overall accuracy of 52.30, significantly lower than GPT-3.5 (67.29) and GPT-4 (74.91). However, when enhanced with RAG, Phi-2's accuracy in the 'Standards Specifications' category increases from 44.27% to 56.63%, approaching GPT-3.5's performance. In a network modeling use case, Phi-2+RAG provides a more accurate energy consumption model than the standalone Phi-2, with a significantly lower mean absolute percentage error (MAPE) of 4.3% compared to 78% for the standalone model. In a user association problem, Phi-2 demonstrates 93% accuracy with two options but shows decreasing accuracy as the number of options increases, highlighting its limitations in complex reasoning tasks. While SLMs like Phi-2 excel in information retrieval and task execution, they struggle with intricate reasoning. The integration of RAG significantly enhances their performance in knowledge-intensive tasks, bridging the gap with larger models. The study concludes that SLMs, when augmented with RAG, can achieve performance comparable to larger models in specific telecom tasks, demonstrating their potential in specialized domains.The increasing interest in Large Language Models (LLMs) in the telecommunications sector highlights their potential to revolutionize operational efficiency. However, their deployment is hindered by their large size and high computational demands, making them unsuitable for resource-constrained environments. Recent advancements have led to the development of small language models (SLMs) that perform comparably to larger models in tasks like coding and common-sense reasoning. Phi-2, a compact yet powerful SLM, is an example of this trend. This paper evaluates Phi-2's understanding of the telecommunications domain and enhances its capabilities using a Retrieval-Augmented Generation (RAG) approach, integrating an extensive knowledge base of telecom standards. The enhanced Phi-2 model demonstrates improved accuracy in answering telecom standard questions, rivaling the performance of GPT-3.5. The paper also explores Phi-2's capabilities in problem-solving scenarios within the telecom sector, highlighting its potential and limitations. The study compares Phi-2 with larger models like GPT-3.5 and GPT-4, using the TeleQnA dataset to assess their telecom knowledge. Phi-2 achieves an overall accuracy of 52.30, significantly lower than GPT-3.5 (67.29) and GPT-4 (74.91). However, when enhanced with RAG, Phi-2's accuracy in the 'Standards Specifications' category increases from 44.27% to 56.63%, approaching GPT-3.5's performance. In a network modeling use case, Phi-2+RAG provides a more accurate energy consumption model than the standalone Phi-2, with a significantly lower mean absolute percentage error (MAPE) of 4.3% compared to 78% for the standalone model. In a user association problem, Phi-2 demonstrates 93% accuracy with two options but shows decreasing accuracy as the number of options increases, highlighting its limitations in complex reasoning tasks. While SLMs like Phi-2 excel in information retrieval and task execution, they struggle with intricate reasoning. The integration of RAG significantly enhances their performance in knowledge-intensive tasks, bridging the gap with larger models. The study concludes that SLMs, when augmented with RAG, can achieve performance comparable to larger models in specific telecom tasks, demonstrating their potential in specialized domains.