Opleiding: Building Large Language Models
The course Building Large Language Models from SpiralTrain teaches you how to design, train, fine-tune, and deploy transformer-based LLMs using PyTorch and modern tooling.
LLM Intro
The course starts by explaining what LLMs are, where they’re used, and the lifecycle of building vs. using them. We introduce the Transformer/GPT architecture, how models learn from large datasets, and when to use classic QA versus RAG.
Working with Text Data
You’ll move from raw text to model-ready tensors: tokenization (e.g., BPE), token→ID mapping, special/context tokens, and sliding-window sampling. We cover embeddings and positional encodings while handling unknown words and basic sentence structure.
Attentions Mechanism
This module demystifies self-attention for long-sequence modeling: queries, keys, values, and causal masking to hide future tokens. We add positional encoding, multi-head attention, and stacked layers to capture dependencies across different parts of the input.
Pytorch Deep Learning
This module explains PyTorch fundamentals—tensors, core operations, and training loops—with the tooling to measure model quality. We cover feature scaling/normalization (including categorical features), activation and loss functions, and backpropagation.
