This paper provides an introduction to Vision-Language Models (VLMs), covering their training paradigms, evaluation methods, and practical considerations. VLMs aim to bridge the gap between vision and language, enabling applications such as image captioning and text-to-image generation. The paper categorizes VLMs into four main training paradigms: contrastive training, masking objectives, generative models, and pre-trained backbones. Each paradigm is explained with examples and recent advancements, such as CLIP, FLAVA, and MaskVLM. The paper also discusses the importance of training data, including data curation, pruning, and diversity, and provides guidance on software, hyperparameters, and model selection. Additionally, it covers techniques for improving grounding, alignment, and text-rich image understanding, as well as parameter-efficient fine-tuning. The evaluation section highlights the challenges and limitations of current benchmarks, emphasizing the need for bias and hallucination measurements. Finally, the paper explores the extension of VLMs to video data, noting the computational and temporal challenges. The goal is to provide a comprehensive guide for researchers and practitioners entering the field of VLMs.This paper provides an introduction to Vision-Language Models (VLMs), covering their training paradigms, evaluation methods, and practical considerations. VLMs aim to bridge the gap between vision and language, enabling applications such as image captioning and text-to-image generation. The paper categorizes VLMs into four main training paradigms: contrastive training, masking objectives, generative models, and pre-trained backbones. Each paradigm is explained with examples and recent advancements, such as CLIP, FLAVA, and MaskVLM. The paper also discusses the importance of training data, including data curation, pruning, and diversity, and provides guidance on software, hyperparameters, and model selection. Additionally, it covers techniques for improving grounding, alignment, and text-rich image understanding, as well as parameter-efficient fine-tuning. The evaluation section highlights the challenges and limitations of current benchmarks, emphasizing the need for bias and hallucination measurements. Finally, the paper explores the extension of VLMs to video data, noting the computational and temporal challenges. The goal is to provide a comprehensive guide for researchers and practitioners entering the field of VLMs.