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
const DL_ERROR_KEYS: Record<string, string> = {
"E-DL-NET": "models.err.net",
"E-DL-SRC": "models.err.src",
"E-DL-DISK": "models.err.disk",
"E-MODEL-HASH": "models.err.hash",
"E-MODEL-INVALID": "models.err.invalid",
"E-MANIFEST-SIG": "models.err.sig",
"E-MANIFEST-EXPIRED": "models.err.expired",
"E-MANIFEST-MINVER": "models.err.minver",
"E-SRV-CONFLICT": "models.err.conflict",
};
export function downloadErrKey(code: string | null): string {
return (code && DL_ERROR_KEYS[code]) || "models.err.generic";
}
export function hasDownloadErr(code: string | null): boolean {
return code !== null && code in DL_ERROR_KEYS;
}