Skip to content

Foundation Models

The full life cycle of language and multimodal models: architecture and training, applications and agents, multimodality and action interfaces, inference, deployment and world models.

Start here

  • Get Started

    A fourteen-day loop: a small GPT, LoRA tuning, retrieval QA with sources, quantised and running offline on an edge device.

    Open

  • Overview

    Terms, system components and how the directions relate.

    Open

  • Roadmap

    Literacy, implementation, use, adaptation, then go deeper by practice.

    Open

Knowledge

  • Architecture & Representation

    How tokens, embeddings and attention make up a model.

    Open

  • Training & Adaptation

    What pre-training, SFT, LoRA and preference optimisation each change.

    Open

  • Applications & Agents

    RAG, tool use and agents, and how to evaluate them.

    Open

  • Multimodality & Action

    How images, language, state and actions connect, up to VLA.

    Open

  • Inference & Deployment

    Quality, latency, throughput and memory; quantisation and KV cache.

    Open

  • GPU & Kernel Development

    Execution model and memory hierarchy, roofline, fusion, FlashAttention and Triton.

    Open

  • World Models

    Predicting how the environment changes, for planning and learning.

    Open

Projects

  • Guided Projects

    F0–F6: from model concepts to a small GPT, retrieval QA and inference comparison.

    Open

  • Project · LLM Quantization

    From RTN to GPTQ / AWQ / SmoothQuant: accuracy, memory and speed.

    Open

Further

  • Artificial Intelligence

    Learning methods and engineering shared across directions.

    Open

  • Embodied Intelligence

    Skill calls, VLA and world models inside the robot loop.

    Open