Publish gemma-4-e2b-it-q4k-ehf16-af32-int4ple registry metadata

registry/catalog.json

@@ -245,6 +245,67 @@
         }
       }
     },
+    {
+      "modelId": "gemma-4-e2b-it-q4k-ehf16-af32-int4ple",
+      "family": "gemma4",
+      "label": "Gemma 4 E2B Instruct (Q4K/F32a/INT4 PLE)",
+      "description": "Gemma 4 E2B multimodal model. Q4K weights, INT4 per-layer embeddings with per-row F32 scales, F16 embeddings/head/vision/audio/projector, F32 accumulation.",
+      "modes": [
+        "text",
+        "vision"
+      ],
+      "aliases": [
+        "gemma4-e2b-int4ple",
+        "gemma-4-e2b-int4ple"
+      ],
+      "artifact": {
+        "format": "rdrr"
+      },
+      "hf": {
+        "repoId": "Clocksmith/rdrr",
+        "path": "models/gemma-4-e2b-it-q4k-ehf16-af32-int4ple",
+        "revision": "d1cd559f4cc1507d48c8d14282628153a5f90979"
+      },
+      "baseUrl": null,
+      "quickstart": false,
+      "sizeBytes": 4009916310,
+      "recommended": false,
+      "sortOrder": 14,
+      "lifecycle": {
+        "availability": {
+          "curated": false,
+          "local": true,
+          "hf": true
+        },
+        "status": {
+          "runtime": "active",
+          "conversion": "ready",
+          "demo": "none",
+          "tested": "verified"
+        },
+        "tested": {
+          "suite": "debug",
+          "surface": [
+            "browser",
+            "node"
+          ],
+          "result": "pass",
+          "lastVerifiedAt": "2026-04-20",
+          "source": "vendor-compare",
+          "notes": "Browser/WebGPU on AMD Strix Halo / Radeon 8060S re-verified 2026-04-20 with release decode cadence. Compare receipts benchmarks/vendors/results/compare_20260420T162233.json (t=0 greedy) and benchmarks/vendors/results/compare_20260420T163016.json (t=1 topK50) show Doppler decode throughput ahead of the paired Transformers.js ONNX/q4f16 runner on warm OPFS. Node/WebGPU on AMD RDNA-3 was previously verified from the repo-local INT4 PLE artifact with deterministic greedy text smokes. Direct LiteRT .task source still loads but remains numerically wrong.",
+          "contracts": {
+            "executionContractOk": true,
+            "layerPatternContractOk": true
+          }
+        }
+      },
+      "vendorBenchmark": {
+        "transformersjs": {
+          "repoId": "onnx-community/gemma-4-E2B-it-ONNX",
+          "dtype": "q4f16"
+        }
+      }
+    },
     {
       "modelId": "translategemma-4b-it-q4k-ehf16-af32",
       "family": "translategemma",