Publish qwen-3-6-27b-q4k-ehaf16 registry metadata

registry/catalog.json

@@ -1,7 +1,7 @@
 {
   "version": 1,
   "lifecycleSchemaVersion": 1,
-  "updatedAt": "2026-04-20",
+  "updatedAt": "2026-04-28",
   "models": [
     {
       "modelId": "gemma-3-270m-it-q4k-ehf16-af32",
@@ -248,9 +248,9 @@
             "node"
           ],
           "result": "pass",
-          "lastVerifiedAt": "2026-04-20",
+          "lastVerifiedAt": "2026-04-21",
           "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_20260420T162522.json (t=0 greedy) and benchmarks/vendors/results/compare_20260420T162831.json (t=1 topK50) show Doppler decode throughput ahead of the paired Transformers.js ONNX/q4f16 runner on warm OPFS. Apple M3 / Node WebGPU darwin execution contract pass remains from reports/gemma-4-e2b-it-q4k-ehf16-af32/2026-04-18T19-17-45.997Z.json. Manifest-only republish to HF 2026-04-18 at revision c690ac0274891afbddb8afb776d000e4c0051b7b.",
+          "notes": "Browser/WebGPU on AMD Strix Halo / Radeon 8060S re-verified 2026-04-21 with compare-owned Gemma 4 prompt rendering. Compare receipt benchmarks/vendors/results/compare_20260421T001902.json shows valid non-empty output and Doppler decode throughput ahead of the paired Transformers.js ONNX/q4f16 runner on warm OPFS (15.49 vs 10.58 tok/s), but exact generated text still mismatches TJS, so this is performance evidence rather than correctness-parity evidence. Apple M3 / Node WebGPU darwin execution contract pass remains from reports/gemma-4-e2b-it-q4k-ehf16-af32/2026-04-18T19-17-45.997Z.json. Full artifact republished to HF 2026-04-21 at revision 2070777c77047d54e6eae105f6dcb1891cf6f21a.",
           "contracts": {
             "executionContractOk": true
           }
@@ -291,7 +291,7 @@
         "revision": "dbb354acb00108965c5324ca7b6fecc198d12501"
       },
       "baseUrl": null,
-      "quickstart": false,
+      "quickstart": true,
       "sizeBytes": 4009916310,
       "recommended": false,
       "sortOrder": 14,
@@ -449,9 +449,9 @@
             "node"
           ],
           "result": "pass",
-          "lastVerifiedAt": "2026-04-20",
+          "lastVerifiedAt": "2026-04-21",
           "source": "vendor-compare",
-          "notes": "Browser/WebGPU on AMD Strix Halo / Radeon 8060S re-verified 2026-04-20 after the quantized tied-LM-head update. Compare receipts benchmarks/vendors/results/compare_20260420T163342.json (t=0 greedy) and benchmarks/vendors/results/compare_20260420T163425.json (t=1 topK50) show Doppler decode throughput ahead of the paired Transformers.js runner on warm OPFS. Earlier 2026-04-18 Node/WebGPU verification after the Qwen prefill kernel-ref swap remains valid; execution contract gate pass. Manifest-only republish to HF 2026-04-18 at revision 1c509e37e4430d7ffd094ce5dfa623792603670e.",
+          "notes": "Browser/WebGPU on AMD Strix Halo / Radeon 8060S re-verified 2026-04-21 from the repaired Clocksmith/rdrr hosted manifest. Compare receipt benchmarks/vendors/results/compare_20260421T002103.json shows exact output match against the paired Transformers.js runner, 64/64 prompt tokens, Doppler decode 61.76 vs TJS 36.14 tok/s, and TTFT 541.0 ms vs 7019.3 ms. Earlier 2026-04-18 Node/WebGPU verification after the Qwen prefill kernel-ref swap remains valid; execution contract gate pass. Full artifact republished to HF 2026-04-21 at revision cf02075803e0a8de7ab26bd76888f55ded35ac5f.",
           "contracts": {
             "executionContractOk": true
           }
@@ -517,9 +517,9 @@
             "node"
           ],
           "result": "pass",
-          "lastVerifiedAt": "2026-04-20",
+          "lastVerifiedAt": "2026-04-21",
           "source": "vendor-compare",
-          "notes": "Browser/WebGPU on AMD Strix Halo / Radeon 8060S re-verified 2026-04-20 after the Qwen 3.5 2B retune; execution contract pass with manifest-owned decodeBatchSize=12. Receipt: reports/qwen-3-5-2b-q4k-ehaf16/2026-04-20T12-49-30.650Z.json. Current execution-v1 uses the fixed fused-Q4 main_gemv transformer decode path, keeps stable fused-Q4 main_multicol declared as fallback, upgrades the tied LM head to Q4 via optimized lm_head_q4 GEMV 64x4, upgrades prefill projections to q4_widetile, and upgrades prefill attention to attn_head256. Compare receipts benchmarks/vendors/results/compare_20260420T163143.json (t=0 greedy) and benchmarks/vendors/results/compare_20260420T163255.json (t=1 topK50, run with --allow-non-comparable-lane because the lane remains capability_only) show Doppler decode throughput ahead of the paired Transformers.js runner on warm OPFS. Manifest-only republish to HF 2026-04-18 at revision 3f54bc54c19181bd737d1773dcd70fb783297e73.",
+          "notes": "Browser/WebGPU on AMD Strix Halo / Radeon 8060S re-verified 2026-04-21 from the repaired Clocksmith/rdrr hosted manifest; execution contract pass with manifest-owned decodeBatchSize=12. Current execution-v1 uses the fixed fused-Q4 main_gemv transformer decode path, keeps stable fused-Q4 main_multicol declared as fallback, upgrades the tied LM head to Q4 via optimized lm_head_q4 GEMV 64x4, upgrades prefill projections to q4_widetile, and upgrades prefill attention to attn_head256. Compare receipt benchmarks/vendors/results/compare_20260421T002238.json shows exact output match against the paired Transformers.js runner, 64/64 prompt tokens, Doppler decode 50.00 vs TJS 32.66 tok/s, and TTFT 602.9 ms vs 7297.8 ms; run with --allow-non-comparable-lane because the lane remains capability_only. Full artifact republished to HF 2026-04-21 at revision a8c45dd885a789042d3b82c95b471d66ca8d5152.",
           "contracts": {
             "executionContractOk": true
           }
@@ -537,6 +537,66 @@
       "artifactCompleteness": "complete",
       "runtimePromotionState": "manifest-owned",
       "weightsRefAllowed": false
+    },
+    {
+      "modelId": "qwen-3-6-27b-q4k-ehaf16",
+      "family": "qwen3",
+      "label": "Qwen 3.6 27B (Q4K)",
+      "description": "Qwen 3.6 27B text model with hybrid linear attention. Q4K weights, F16 embeddings, F32 accumulation.",
+      "modes": [
+        "text"
+      ],
+      "aliases": [
+        "qwen3.6-27b",
+        "qwen-3-6-27b",
+        "Qwen/Qwen3.6-27B"
+      ],
+      "artifact": {
+        "format": "rdrr"
+      },
+      "hf": {
+        "repoId": "Clocksmith/rdrr",
+        "path": "models/qwen-3-6-27b-q4k-ehaf16",
+        "revision": "2bc4bfd905fe214303545a00db61031d4088b379"
+      },
+      "baseUrl": null,
+      "quickstart": false,
+      "sizeBytes": 16994352128,
+      "recommended": false,
+      "sortOrder": 22,
+      "lifecycle": {
+        "availability": {
+          "curated": false,
+          "local": true,
+          "hf": true
+        },
+        "status": {
+          "runtime": "active",
+          "conversion": "ready",
+          "demo": "none",
+          "tested": "verified"
+        },
+        "tested": {
+          "suite": "inference",
+          "surface": [
+            "browser"
+          ],
+          "result": "pass",
+          "lastVerifiedAt": "2026-04-28",
+          "source": "browser-webgpu-smoke",
+          "notes": "Local browser/WebGPU smoke report reports/qwen-3-6-27b-q4k-ehaf16/2026-04-28T01-19-10.624Z.json captured a 4-token deterministic reference transcript and executionContractOk=true. Hosted artifact only; quickstart and demo download remain disabled.",
+          "contracts": {
+            "executionContractOk": true
+          }
+        }
+      },
+      "vendorBenchmark": null,
+      "sourceCheckpointId": "Qwen/Qwen3.6-27B",
+      "weightPackId": "qwen-3-6-27b-q4k-ehaf16-wp-catalog-v1",
+      "manifestVariantId": "qwen-3-6-27b-q4k-ehaf16-mv-exec-v1",
+      "artifactCompleteness": "complete",
+      "runtimePromotionState": "manifest-owned",
+      "weightsRefAllowed": false
     }
   ]
 }