Upload holo-load2bit.mjs with huggingface_hub

holo-load2bit.mjs

@@ -114,8 +114,12 @@ export async function loadKappaObject(baseUrl, opts = {}) {
       // digest disagrees with the pin, re-derive on the CPU — so a GPU quirk degrades to CPU speed, never fails
       // the load. Only a CPU mismatch (real corruption) throws. Set globalThis.__gpuVerify=false to force CPU.
       if (gpuBlake3Available() && (typeof globalThis === "undefined" || globalThis.__gpuVerify !== false)) {
-        try { got = "blake3:" + await gpuBlake3Hex(gz); } catch (e) { got = "blake3:" + blake3hex(gz); }
-        if (got !== kappa) got = "blake3:" + blake3hex(gz);   // GPU disagreed → trust the CPU verifier
+        const t0 = performance.now();
+        let h = null; try { h = await gpuBlake3Hex(gz); } catch (e) {}
+        if (h && "blake3:" + h === kappa) {                   // verified ON THE GPU, matches the pin
+          got = kappa;
+          try { const v = globalThis.__vs || (globalThis.__vs = { n: 0, bytes: 0, ms: 0 }); v.n++; v.bytes += gz.length; v.ms += performance.now() - t0; } catch (e) {}
+        } else got = "blake3:" + blake3hex(gz);               // GPU errored/disagreed → CPU re-derive (never fail on a GPU quirk)
       } else got = "blake3:" + blake3hex(gz);
     } else got = "sha256:" + hex(await crypto.subtle.digest("SHA-256", gz));
     if (got !== kappa) throw new Error("κ MISMATCH " + kappa.slice(0, 24));