3 Generative AI Models and LLM: Training Techniques and Evaluation Metrics
-
C. Arun
Abstract
Generative artificial intelligence (AI) has been a prominent technique across data-driven applications, which uses deep learning architecture to learn the underlying characteristic of the sample to build the knowledge base in generating synthetic samples that mimic the real distribution. Generative AI models are ideal solutions where models suffer due to scarcity of data sample that hinders the training process be it text, video, audio, and image. Training the model plays a pivotal role, where it discovers the hidden pattern and understands the intrinsic behavior of samples that aid in generating realistic samples. The volume of data that is available for training and the computing power required pose threat on the performance of the intelligent systems, where large language models (LLM) has been an ideal solution. LLMs are generative AI systems that understand human language and provide intelligent, creative solutions to questions. Complex architecture of LLM allows them to capture the intricacies of language more precise, enabling to generate coherent and contextually relevant outputs. This chapter delves into comprehensive analysis on the well-known generative AI models such as generative adversarial networks, transformers, and LangChain. Generative AI employs different training techniques such as reinforcement learning, adversarial training, variational inference, transfer learning, and progressive training on diverse application domains. Furthermore, the study examines the crucial aspect of evaluating the effectiveness of generative models, using a variety of metrics ranging from BLUE, inception score, perplexity, Frechet inception distance, precision, ROUGE, recall, METEOR, BERT, MoverScore, and many more. A comparative analysis of these metrics offers insights into their respective advantages and disadvantages, aiding practitioners and researchers in selecting benchmarks that align with their specific use cases.
Abstract
Generative artificial intelligence (AI) has been a prominent technique across data-driven applications, which uses deep learning architecture to learn the underlying characteristic of the sample to build the knowledge base in generating synthetic samples that mimic the real distribution. Generative AI models are ideal solutions where models suffer due to scarcity of data sample that hinders the training process be it text, video, audio, and image. Training the model plays a pivotal role, where it discovers the hidden pattern and understands the intrinsic behavior of samples that aid in generating realistic samples. The volume of data that is available for training and the computing power required pose threat on the performance of the intelligent systems, where large language models (LLM) has been an ideal solution. LLMs are generative AI systems that understand human language and provide intelligent, creative solutions to questions. Complex architecture of LLM allows them to capture the intricacies of language more precise, enabling to generate coherent and contextually relevant outputs. This chapter delves into comprehensive analysis on the well-known generative AI models such as generative adversarial networks, transformers, and LangChain. Generative AI employs different training techniques such as reinforcement learning, adversarial training, variational inference, transfer learning, and progressive training on diverse application domains. Furthermore, the study examines the crucial aspect of evaluating the effectiveness of generative models, using a variety of metrics ranging from BLUE, inception score, perplexity, Frechet inception distance, precision, ROUGE, recall, METEOR, BERT, MoverScore, and many more. A comparative analysis of these metrics offers insights into their respective advantages and disadvantages, aiding practitioners and researchers in selecting benchmarks that align with their specific use cases.
Chapters in this book
- Frontmatter I
- Preface V
- Contents VII
- About the Editors IX
- List of Contributors XI
- 1 Unveiling the Power of Generative AI: A Journey into Large Language Models 1
- 2 Early Roots of Generative AI Models and LLM: A Diverse Landscape 23
- 3 Generative AI Models and LLM: Training Techniques and Evaluation Metrics 43
- 4 Importance of Prompt Engineering in Generative AI Models 69
- 5 LLM Pretraining Methods 93
- 6 LLM Fine-Tuning: Instruction and Parameter-Efficient Fine-Tuning (PEFT) 117
- 7 Reinforcement Learning from Human Feedback (RLHF) 135
- 8 Exploring the Applications on Generative AI and LLM 155
- 9 Bias and Fairness in Generative AI 177
- 10 Future Directions and Open Problems in Generative AI 193
- 11 Optimizing Sustainable Project Management Life Cycle Using Generative AI Modeling 213
- 12 Generative AI and LLM: Case Study in Finance 231
- 13 Generative AI and LLM: Case Study in E-Commerce 253
- Index 273
Chapters in this book
- Frontmatter I
- Preface V
- Contents VII
- About the Editors IX
- List of Contributors XI
- 1 Unveiling the Power of Generative AI: A Journey into Large Language Models 1
- 2 Early Roots of Generative AI Models and LLM: A Diverse Landscape 23
- 3 Generative AI Models and LLM: Training Techniques and Evaluation Metrics 43
- 4 Importance of Prompt Engineering in Generative AI Models 69
- 5 LLM Pretraining Methods 93
- 6 LLM Fine-Tuning: Instruction and Parameter-Efficient Fine-Tuning (PEFT) 117
- 7 Reinforcement Learning from Human Feedback (RLHF) 135
- 8 Exploring the Applications on Generative AI and LLM 155
- 9 Bias and Fairness in Generative AI 177
- 10 Future Directions and Open Problems in Generative AI 193
- 11 Optimizing Sustainable Project Management Life Cycle Using Generative AI Modeling 213
- 12 Generative AI and LLM: Case Study in Finance 231
- 13 Generative AI and LLM: Case Study in E-Commerce 253
- Index 273
