
Already know some of this? A 2-minute test skips the lessons you know.
Take the placement testA free 11-lesson course built from freeCodeCamp.org's 3h 29m of YouTube video. 3 days at 90 minutes a day, with study notes and a quiz in every lesson. A 2-minute placement test skips what you already know.
Taught by freeCodeCamp.org on YouTube. Videos play in YouTube's own player, so the creator keeps every view.
Understand the full LLM training pipeline and learn to export and filter WhatsApp chat data for model training.
Apply regex filters to clean chat exports, concatenate messages into a single text sequence, and grasp the theory of Byte Pair Encoding for tokenization.
Train a configurable BPE tokenizer using the minbpe library, inspect the vocabulary, and add special tokens required for fine-tuning.
Learn the components of a decoder-only transformer (embeddings, self-attention, MLP, layer norm) and the key hyperparameters that control model size and context length.
Build the GPT-style transformer from scratch in PyTorch: head, multi-head attention, block, and model classes with weight tying and generation logic.
Set up padded data loaders, implement a training loop with periodic validation loss estimation, and save checkpoints to resume interrupted runs.
Diagnose underfitting from loss curves, then systematically increase embedding size, layers, and batch strategy to improve model capacity.
Structure chat logs into system/user/assistant turns, combine pairs into fixed-length padded sequences, and save a fine-tuning dataset with end-of-text tokens.
Load a pre-trained checkpoint, fine-tune with a lower learning rate using a while-loop generator that stops on the end-of-text token, and evaluate sample outputs.
Merge multiple conversation turns into single sequences up to block size, apply loss masking on user/system tokens, and fine-tune for multi-turn dialogue.
Freeze base model weights, inject trainable LoRA matrices into linear layers, and fine-tune with drastically fewer parameters.