LLM Model Notes
Updates
New pages and substantive revisions are recorded here. Minor copy and styling corrections are not listed.
Follow updates via Atom for new and materially revised notes.
Added three notes: reading an open-weight model card, serving mixture-of-experts models, and measuring tokenizer efficiency for Chinese text. Published the model tracker, the glossary, and an archive of all notes by topic, and started the automatically refreshed news page.
Added three connected guides: reading benchmark results carefully, measuring latency, throughput, and context, and building repeatable model comparisons. Expanded the site scope and shared methodology.
Published the project overview, comparison methodology, and introductory notes on benchmark interpretation and inference trade-offs.