Generative AI : LLM, Fine-tuning, RAG & Prompt engineering

Explore the evolution from rule-based systems to transformers and large language models, and see how self-attention, billions of parameters, and data-rich training teach these models to generate human-like text.

Explore how large language models train by using text data from books, journals, and websites. Predict the next word, refining parameters and weights through feedback to generate coherent text.

Identify how model architecture and dataset quality and diversity shape LLM performance, and optimize tasks using zero-shot, few-shot, and domain adaptation; unlock real-world applications.

Explore the wide range of large language model use cases, from content generation and translation to text summarization, question answering in chatbots, content moderation, information retrieval, and educational tools.

Explore the challenges facing large language models, including data bias and training data quality, hallucinations, and ethical, legal, and privacy concerns. Assess deployment hurdles from scalability to latency and security.

Master domain-specific fine-tuning by adapting a pre-trained model with a specialized dataset, training further, and optimizing for domain tasks. Learn benefits like specialization, resource efficiency, and improved domain performance.

Tailor large language models to specific domains through domain-specific pre-training and fine-tuning. Bloomberg GPT demonstrates finance-focused tasks like financial sentiment analysis, named entity recognition, and financial question answering.

Choose domain-specific pretraining for domains with ample data and architecture customization, or apply domain-specific fine tuning and retrieval augmented generation to maintain up-to-date results and reduce hallucinations.

tailor domain-specific llms to understand the nuances and terminology of fields like healthcare, finance, and law, reducing hallucinations and errors and improving reliability for specialized content.

Domain-specific LLMs deepen understanding of specialized terminology, overcome general models' limitations, improve privacy compliance, and boost productivity through tailored, accurate, and contextual interactions.

Explore how to craft and refine prompts for language models, applying chain of thought and tree of thought methods, transfer learning, and bias reduction to improve safe, relevant outputs.

Explore the four core prompt components—instruction, context, input data, and output indicator—and learn advanced techniques like chain-of-thought, step-by-step prompts, and graph of thought for robust reasoning and verification.

Learn automatic prompt engineering with an app that generates diverse instruction candidates, scores them using a function, and selects the best prompt, while self-consistency and react enable external tools reasoning.

Explore prompt engineering with the Mistral AI API through a live project, learning to summarize, answer questions, and classify text while building and testing API-driven prompts.

Explore fine tuning and retrieval augmented generation (rag) to boost large language model performance on domain-specific tasks, using data ingestion, retrieval, and synthesis, with medical and legal examples.

examine unsupervised and supervised fine-tuning methods for large language models, including full and partial approaches, contrastive learning, and parameter-efficient techniques such as low rank adaptation.

Explore instruction fine tuning to improve generalization by adding explicit instructions to input-output pairs. Learn about the Natural Instructions Dataset, its tasks, and applications in translation, summarization, and question answering.

Explore reinforcement learning from human feedback to align LLMs with human values by combining pre-training, reward-model training, and fine-tuning with reinforcement learning using PPO, guided by human preferences.

direct preference optimization (DPO) directly incorporates user preferences by comparing model outputs and adjusting parameters, without a reward model, offering a simpler, faster alternative to RLHF and maximum likelihood training.

Explore parameter efficient fine tuning (FFT) to cut computational and memory demands by updating only a subset of parameters, enabling adaptation across NLP, CV, and audio.

Explore retrieval augmented generation with Llama 2, embedding queries, retrieving top-k documents, and synthesizing context using a rag pipeline; then fine-tune on limited hardware with quantized low rank adaptation.

Explore advanced retrieval augmented generation techniques, including data ingestion, embedding, and retrieval strategies, to fine-tune Rag models and balance parametric and non-parametric memory.

Discover how retrieval augmented generation (rag) enhances large language models by integrating external information, reducing hallucinations and enabling source verification, through a three-part pipeline: ingestion, retrieval, synthesis.

Explore improvements to the ingestion component of RAG systems by applying advanced chunking and indexing strategies, including content-based, sentence-level, and recursive chunking, plus various indexing approaches.

Explore retrieval techniques for rag, including hypothetical questions, context enrichment, sentence window retrieval, auto merging, fusion retrieval, reranking filtering, and routing to improve accuracy.

Learn advanced generation in retrieval augmented generation, including response synthesis, iterative refinement, summarization, concatenation, and multiple answers, plus encoder and LLM fine tuning and joint tuning for improved accuracy.

Explore the retrieval augmented generation (RAG) pipeline, contrasting with pure generation, and learn how RAG lets LLMs use new confidential data like HR policy and financial analytics.

Explore how retrieval augmented generation strengthens large language models by ingesting external documents, embedding them into a vector database, and performing semantic retrieval before generation.

Build a retrieval augmented generation system with Lang Chain and Gemini Pro, loading a pdf, creating embeddings, storing them in a Chroma vector store, and building a retrieval-based QA chain.

Evaluate lms across pipeline and model dimensions, covering prompts, retrieval, and context quality. Build and deploy lms using retrieval augmented generation, memory, agents, fine-tuning, and lm ops on Hugging Face.

Explore the tool ecosystem for building and deploying LLM applications. Learn input processing, retrieval augmented generation, vector databases, orchestration, hosting, and evaluation.

Explore how proprietary and open-source LMS, RAG, fine-tuning, prompt engineering, and multi-LLM orchestration power an AI-based customer support agent.

Identify business needs and define AI outcomes to set scope and roi. Prepare data, choose models, deploy securely, build user-centric applications, and assess accuracy and ethics.