mlabonne/llm-course

Features

• Large Language Models - [New model release!](https://mlabonne.github.io/blog/posts/2026-02-12_GLM-5.html) Large Language Models
• Retrieval Augmented Generation Systems - Building intelligent systems that ground model responses in external data sources to improve accuracy and reduce hallucinations in production.
• Transformer Architectures - Processes input tokens through stacked attention layers to capture long-range dependencies and contextual relationships within sequential data.
• Supervised Fine-Tuning - SFT turns base models into helpful assistants, capable of answering questions and following instructions. During this process, they learn how to structure answers and reactivate a subset of knowledge learned during pre-t
• Preference-Based Model Alignments - Refines pre-trained model behavior by optimizing against human-provided feedback or reward signals to ensure helpful and safe outputs.
• Fine-Tuning Strategies - Mastering the end-to-end process of adapting pre-trained models through supervised fine-tuning and preference alignment techniques for specific tasks.
• Supervised Instruction Fine-Tuning - Adapts base models to specific task formats by training on curated datasets containing structured input-output instruction pairs.
• Retrieval Pipelines - Augments model outputs by dynamically fetching relevant external documents from a vector database to ground generated responses.
• Inference Optimization Techniques - Implementing techniques like quantization and efficient attention mechanisms to reduce hardware requirements and maximize throughput for deployed models.
• Agentic Orchestration Frameworks - Enables autonomous task execution by allowing models to reason about environment states and invoke external APIs or functions.
• Neural Networks - Neural networks are a fundamental part of many machine learning models, particularly in the realm of deep learning. To utilize them effectively, a comprehensive understanding of their design and mechanics is essential. -
• Post-Training Datasets - Post-training datasets have a precise structure with instructions and answers (supervised fine-tuning) or instructions and chosen/rejected answers (preference alignment). Conversational structures are a lot rarer than th
• Preference Alignment Strategies - Preference alignment is a second stage in the post-training pipeline, focused on aligning generated answers with human preferences. This stage was designed to tune the tone of LLMs and reduce toxicity and hallucinations.
• Model Merging Strategies - Combining multiple specialized models into a single, more capable architecture using advanced merging algorithms and mixture-of-experts strategies.
• Quantization Methods - Reduces memory footprint and computational overhead by mapping high-precision floating-point weights to lower-bit integer representations.
• Weight-Space Merging Techniques - Combines multiple fine-tuned model checkpoints into a single unified architecture without additional training to aggregate diverse capabilities.
• Model Evaluation Frameworks - Reliably evaluating LLMs is a complex but essential task guiding data generation and training. It provides invaluable feedback about areas of improvement, which can be leveraged to modify the data mixture, quality, and t
• Autonomous Agent Frameworks - Designing systems where models reason about environments and utilize external tools to perform complex, multi-step tasks without human intervention.
• Agentic Reasoning Frameworks - An LLM agent can autonomously perform tasks by taking actions based on reasoning about its environment, typically through the use of tools or functions to interact with external systems. * **Agent fundamentals**: Agents
• RAG Orchestration Patterns - With RAG, LLMs retrieve contextual documents from a database to improve the accuracy of their answers. RAG is a popular way of augmenting the model's knowledge without any fine-tuning. * **Orchestrators**: Orchestrators
• Model Merging Techniques - [Agent swarm and early fusion for better vision capabilities](https://mlabonne.github.io/blog/posts/2026-02-19_kimik25.html) Large Language Models
• LLM Engineering Guides - A practical guide detailing the implementation of model training, fine-tuning, deployment, and optimization strategies for production-grade artificial intelligence.
• Fine-Tuning Tutorials - | Notebook | Description | Article | Notebook | |---------------------------------------|-------------------------------------------------------------------------|---------------------------------------------------------
• Educational Curriculum Repositories - A structured collection of learning materials and technical guides covering the end-to-end lifecycle of modern large language models.
• Large Language Model Tutorials - [Agent swarm and early fusion for better vision capabilities](https://mlabonne.github.io/blog/posts/2026-02-19_kimik25.html) Large Language Models
• Local Model Execution - Running LLMs can be difficult due to high hardware requirements. Depending on your use case, you might want to simply consume a model through an API (like GPT-4) or run it locally. In any case, additional prompting and g
• Large Language Model Architectures - An in-depth knowledge of the Transformer architecture is not required, but it's important to understand the main steps of modern LLMs: converting text into numbers through tokenization, processing these tokens through la
• Machine Learning Mathematics - Before mastering machine learning, it is important to understand the fundamental mathematical concepts that power these algorithms. - **Linear Algebra**: This is crucial for understanding many algorithms, especially thos
• Retrieval Augmented Generation Guides - Real-life applications can require complex pipelines, including SQL or graph databases, as well as automatically selecting relevant tools and APIs. These advanced techniques can improve a baseline solution and provide ad
• AI Research Repositories - A curated repository of state-of-the-art methodologies, architectural insights, and experimental trends in the rapidly evolving field of generative models.
• Machine Learning Knowledge Bases - A comprehensive reference for fundamental concepts, mathematical foundations, and advanced techniques in deep learning and natural language processing.
• LLM Security - In addition to traditional security problems associated with software, LLMs have unique weaknesses due to the way they are trained and prompted. * **Prompt hacking**: Different techniques related to prompt engineering, i