Skip to content

Latest commit

 

History

History
17 lines (14 loc) · 1.9 KB

File metadata and controls

17 lines (14 loc) · 1.9 KB

InferenceX Docs

Design rationale and non-obvious conventions. See CLAUDE.md for the quick-start project guide.

Docs

  • Architecture — Why client-first, hash routing, URL state, provider nesting, server-side caching (unstable_cache + blob), in-memory client cache, color system, analytics enforcement
  • D3 Charts — Why 4 effects, in-place mutation, refs for zoom, rAF throttling, HTML tooltips, Pareto directions, gradient labels
  • Data Pipeline — DB schema reasoning, ETL design, transform pipeline, spline method choice, normalizer resolution order (model/GPU/framework)
  • Pitfalls — Failure modes: token type consistency, schema evolution, empty objects, zoom loss, stale closures, disaggregated metrics, negative splines, date stamping, ref stability, cost inheritance
  • GPU Specs — Unit conventions, topology invariants, SVG layout rationale, hardware gotchas
  • TCO Calculator — Why interpolation, composite keys, cost matrix, token type bugs, badge logic, state design
  • Adding Entities — Step-by-step checklists for adding new models, GPUs, precisions, sequences, frameworks (ingest + constants + frontend), plus featuring a day-0 model (launch banner, modal, Quick Comparisons preset)
  • Testing — Requirements, quality standards, pre-commit checklist
  • Data Transforms — Full pipeline from BenchmarkRow to RenderableGraph: type hierarchy, hardware key construction, derived metrics, memoization strategy
  • State Ownership — Which context owns which state, availability filtering cascade, comparison date mechanics, URL param sync
  • Blog — MDX content system, SEO features (OG images, RSS, llms.txt, JSON-LD), TOC sidebar, reading progress, heading links, analytics events