🧠 Model Architectures β€” A Visual History

Comprehensive documentation of LLM architecture evolution across generations. Each model family is traced from version 1 to the latest release with architecture diagrams, benchmark tables, and community perspectives β€” all sourced from official technical papers.

GitHub Pages License: MIT


πŸ“š Model Families

# Model Family Versions Covered Status
01 Qwen 1 β†’ 1.5 β†’ 2 β†’ 2.5 β†’ 3 βœ… Complete
02 Llama 1 β†’ 2 β†’ 3 β†’ 3.1 β†’ 3.2 β†’ 3.3 β†’ 4 βœ… Complete
03 DeepSeek V1 β†’ V2 β†’ V3 β†’ R1 πŸ”œ Planned
04 Gemma 1 β†’ 2 β†’ 3 β†’ 4 βœ… Complete
05 Mistral 7B β†’ Mixtral β†’ Large πŸ”œ Planned
06 Phi 1 β†’ 1.5 β†’ 2 β†’ 3 β†’ 4 πŸ”œ Planned
07 Alpamayo R1 β†’ 1.5 (VLA) βœ… Complete

🎯 What’s Inside Each Document

Every model family document includes:

  • Release timeline with dates, paper links, and parameter counts
  • Cross-version benchmark tables sourced from official papers
  • HTML+CSS architecture diagrams showing what changed between versions
  • 10-point summaries per version covering novel ideas and design decisions
  • Model variant tables (base, instruct, coder, math, MoE sizes)
  • Community perspectives and industry reception
  • Full reference tables linking to all papers, blogs, and repos