Back to Module 1.10: TransformersIn Progress
AI Lesson & Submodule
Decoder-only Architecture
Analyze GPT/Llama models and causal decoding loops.
Why This Matters
Modern generative LLMs are almost exclusively decoder-only, utilizing causal masking to generate text.
What You Will Learn
- •Map autoregressive loops
- •Verify causal splits
- •Describe output logits projection
Concepts Covered
Autoregressive decoding loopCausal splits filtersOutput logits projection matrix
Mapped Foundation Project: Mini Transformer Block Explainer
Visual deconstruction of a standard decoder block, outlining normalizations, skip links, and output projections.
Architecture Preview
Block-by-block diagram tracking vector changes as inputs pass through decoder normalizations and linear mappings.
Input TokensLayer Norm Layer 1Multi-Head Attention Block
Tech Stack Planned
ReactTypeScriptFramer Motion
GitHub: Coming SoonLive Demo: Coming Soon
Coming SoonTechnical Interview Value
- ?Explain how a decoder-only LLM generates text one token at a time autoregressively