@whats-my-model/core
v0.2.1
Published
Pure hardware-fit engine for local GGUF models: will it fit this machine?
Maintainers
Readme
@whats-my-model/core
Pure, dependency-free JavaScript. No DOM, framework, backend, or Tauri.
API
classifyModel(sizeBytes, resources)— will a model of this file size fit? Returns{ tier: "ok" | "tight" | "over" | "unknown", sizeBytes, need, ceiling, target }. Binds against VRAM when a GPU is present (the display-freeze constraint), else system RAM.viabilityLabel/viabilityColor/pressureTier/formatBytes— presentation helpers over the classification.mergeLocalModels(discoveredPaths, manualModels)— turn a scanned folder of.gguffiles into deduped, uniquely-named local-model entries (+baseName/parentDirName/modelStem).- Hardware providers —
browserHardwareProvider()(best-effort, no backend, no VRAM) andmanualHardwareProvider(profile). Hosts with an accurate probe inject their own{ inspect(): Promise<HardwareProfile> }.
resources / HardwareProfile shape: { gpu: { total, used, free, name? } | null, ram: { total, available, used } | null }, all in bytes.
Test
npm test # node --test, zero dependencies