Run 35B MoE models in ~2.5 GB of RAM. Weights stream from SSD; a trained router predicts loads a step ahead. Mac, iPhone, Android, Windows supported
Python
import type { Event, SystemSnapshot } from "../gen";
export type ModelLight = "resident" | "loading" | "installed" | "absent";
const MODEL_EVENT_TYPES = [
"model.load.started",
"model.load.ready",
"model.load.failed",
"model.unloaded",
] as const;
export function latestModelEvent(
store: { latestFolded(type: string, subject: string): Event | undefined },
modelId: string,
): Event | null {
let best: Event | null = null;
for (const t of MODEL_EVENT_TYPES) {
const e = store.latestFolded(t, modelId);
if (e && (!best || e.seq > best.seq)) best = e;
}
return best;
}
export function deriveLight(
snapshot: SystemSnapshot | null,
modelId: string,
latest: Event | null,
): ModelLight {
if (!snapshot) return "absent";
const entry = snapshot.models.find((m) => m.id === modelId);
if (!entry) return "absent";
if (latest && latest.subject === modelId && latest.type === "model.load.started") {
if (entry.state !== "resident") return "loading";
}
return entry.state === "resident" ? "resident" : "installed";
}
export function candidateModels(snapshot: SystemSnapshot | null): string[] {
return (snapshot?.models ?? []).map((m) => m.id);
}