> Dismantling Transformers

Interactive exploration of the Transformer architecture

>Page 1: ITV[--:--:--]
Interactive Transformer Visualizer

Explore the core "Attention Is All You Need" architecture with mathematically accurate visualizations.

●Self-Attention Heatmaps
●Positional Encoding
●Q, K, V Matrices
●Encoder/Decoder Stack
> Deep Dive →
>Page 2: ITT[--:--:--]
Interactive Text Tokenizer

Compare tokenization across BERT, GPT, and T5 with live embedding visualization.

●WordPiece (BERT)
●BPE (GPT)
●SentencePiece (T5)
●Embedding Pipeline
> Tokenize Text →
>Page 3: IPC[--:--:--]
Interactive Path Comparator

Compare BERT, GPT, and T5 with side-by-side architecture and attention pattern visualizations.

●BERT: Encoder-Only
●GPT: Decoder-Only
●T5: Encoder-Decoder
●Pre-training Tasks
> Compare Models →
>Page 4: LDI[--:--:--]
LLM Deconstructor & Integrator

Interactive PyTorch-style code walkthrough with synchronized math and explanations (CDX feature).

●BPE Tokenization
●Embedding Layers
●Attention Pipeline
●Model-Specific Masking
> Code Walkthrough →
ITV→ITT→IPC→LDI
Recommended learning path: Architecture → Tokenization → Model Comparison → Implementation