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 { createRootRoute, createRoute, createRouter, redirect } from "@tanstack/react-router";
import { AppShell } from "./components/AppShell";
import { ChatPage } from "./routes/chat";
import { ModelsPage } from "./routes/models";
import { ServicePage } from "./routes/service";
import { SettingsPage } from "./routes/settings";
const rootRoute = createRootRoute({
component: AppShell,
});
const chatRoute = createRoute({
getParentRoute: () => rootRoute,
path: "/",
component: ChatPage,
});
const modelsRoute = createRoute({
getParentRoute: () => rootRoute,
path: "/models",
component: ModelsPage,
});
const settingsRoute = createRoute({
getParentRoute: () => rootRoute,
path: "/settings",
component: SettingsPage,
});
const serviceRoute = createRoute({
getParentRoute: () => rootRoute,
path: "/service",
component: ServicePage,
});
const legacyChatRedirect = createRoute({
getParentRoute: () => rootRoute,
path: "/chat",
beforeLoad: () => {
throw redirect({ to: "/" });
},
});
const routeTree = rootRoute.addChildren([
chatRoute,
modelsRoute,
legacyChatRedirect,
settingsRoute,
serviceRoute,
]);
export const router = createRouter({ routeTree });
declare module "@tanstack/react-router" {
interface Register {
router: typeof router;
}
}