How Open-Source LLMs Differ: Comparing 130+ Model Architectures
Post · https://dev.to/amananandrai/how-open-source-llms-differ-comparing-130-model-architectures-3b9d
A few weeks ago I wrote about LLM Explained, a free, open-source course of 43 interactive pages that...
First seen by Oriel on .
AI summary
The Architecture Atlas is an open-source interactive reference that explains and compares 130+ LLM architectures through configurable diagrams, matrices, and a searchable gallery.
- It presents transformer evolution from the 2017 encoder–decoder to modern decoder-only blocks using RMSNorm, RoPE, GQA, KV caching, and SwiGLU or MoE layers.
- Fourteen model configurations generate diagrams, counters, comparison rows, and specification cards; all reproduce published parameter sizes within 0.3%, except rounded Kimi K2.
- The comparison matrix shows universal decoder-only pre-norm RMSNorm, RoPE, and gated MLP usage, while GQA, MLA, MoE, and KV-cache sizes vary substantially.
- A gallery adds 120 open-weight architectures with searchable layer strips, attention types, model facts, licences, and links to relevant educational modules.
- The site uses plain JavaScript without frameworks or a build step, and lazy-loads an 86 KB gallery file only when the atlas opens.
#llm architectures #transformers #mixture of experts #web development
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Reactions
Total public reactions reported by the source; distinct from points and likes.
- DEV Community: 9, latest observation ; 0 over observed period since
Comments
Comment count reported by the source.
- DEV Community: 1, latest observation ; 0 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| DEV Community | 5029079 | https://dev.to/amananandrai/how-open-source-llms-differ-comparing-130-model-architectures-3b9d Read discussion |