web-katrain

Browser-based KaTrain-style UI with in-browser KataGo analysis using TensorFlow.js (WebGPU with WASM/CPU fallback).

How it works

• React + Zustand manage game state, settings, and the move tree.
• A Web Worker runs KataGo-style evaluation/search on the current position.
• Weights load from public/models/ or a URL/upload set in Settings.
• Analysis results feed the UI (policy, ownership, winrate/score graphs).

For detailed architecture diagrams and flow explanations, see docs/diagram.md.

Development

• Install: npm install
• Run: npm run dev
• Test: npm test
• Lint: npm run lint
• Build: npm run build then npm run preview

Models

Models live in public/models/. Settings lets you swap URLs or upload weights for the session. Optional parity assets can be pulled from sibling checkouts:

• ../katrain-ref/ – KaTrain reference (Python/Kivy)
• ../KataGo/ – KataGo reference (C++)

Scripts like scripts/fetch-katago-small-model.mjs will copy KaTrain’s default model from ../katrain-ref/katrain/models/ when present.

Performance

For threaded WASM (XNNPACK), serve with COOP/COEP headers to enable SharedArrayBuffer: Cross-Origin-Opener-Policy: same-origin and Cross-Origin-Embedder-Policy: require-corp. Vite dev/preview already sends these headers. The production build includes a public/_headers file for static hosts that honor it (Netlify/Cloudflare Pages); other hosts should set equivalent headers.