clocksmith commited on
Commit
07357a6
·
verified ·
1 Parent(s): f86fe24

Publish qwen-3-reranker-0-6b-q4k-ehf16-af32 registry metadata

Browse files
Files changed (1) hide show
  1. registry/catalog.json +76 -0
registry/catalog.json CHANGED
@@ -483,6 +483,82 @@
483
  "runtimePromotionState": "manifest-owned",
484
  "weightsRefAllowed": false
485
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
486
  {
487
  "modelId": "qwen-3-5-0-8b-q4k-ehaf16",
488
  "family": "qwen3",
 
483
  "runtimePromotionState": "manifest-owned",
484
  "weightsRefAllowed": false
485
  },
486
+ {
487
+ "modelId": "qwen-3-reranker-0-6b-q4k-ehf16-af32",
488
+ "family": "qwen3",
489
+ "label": "Qwen3 Reranker 0.6B (Q4K/F32a)",
490
+ "description": "Qwen3 0.6B SentenceTransformers CrossEncoder reranker target with Q4K attention, F16 MLP projections, and manifest-owned true-logit scoring. The local lane passes semantic rerank verification and benchmark checks; HF publication remains disabled until explicit promotion.",
491
+ "modes": [
492
+ "rerank"
493
+ ],
494
+ "aliases": [
495
+ "qwen3-reranker-0.6b-q4k",
496
+ "qwen3-reranker-0-6b-q4k",
497
+ "Qwen/Qwen3-Reranker-0.6B:q4k"
498
+ ],
499
+ "artifact": {
500
+ "format": "rdrr"
501
+ },
502
+ "hf": {
503
+ "repoId": "Clocksmith/rdrr",
504
+ "revision": "f86fe245b9bbc275cd69af46b1d45d47ea685a55",
505
+ "path": "models/qwen-3-reranker-0-6b-q4k-ehf16-af32"
506
+ },
507
+ "baseUrl": null,
508
+ "quickstart": false,
509
+ "demoVisible": false,
510
+ "sizeBytes": 938321920,
511
+ "recommended": false,
512
+ "sortOrder": 19.1,
513
+ "lifecycle": {
514
+ "availability": {
515
+ "curated": false,
516
+ "local": true,
517
+ "hf": true
518
+ },
519
+ "status": {
520
+ "runtime": "active",
521
+ "conversion": "ready",
522
+ "demo": "none",
523
+ "tested": "verified"
524
+ },
525
+ "tested": {
526
+ "suite": "rerank",
527
+ "surface": [
528
+ "node"
529
+ ],
530
+ "result": "pass",
531
+ "lastVerifiedAt": "2026-07-04",
532
+ "source": "runtime-verify",
533
+ "contracts": {
534
+ "executionContractOk": true
535
+ },
536
+ "notes": "Release evidence: reports/release-claims/qwen-3-reranker-0-6b-q4k-ehf16-af32/2026-07-04T18-42-51.669Z.json. Semantic pair accuracy 1.0 (5/5), WebGPU query topDocumentIndex 0, rerankMs 1725.75 for the three-document smoke request. Speed evidence: reports/release-claims/qwen-3-reranker-0-6b-q4k-ehf16-af32/2026-07-04T18-43-20.530Z.bench.json (medianRerankMs 1546.27 for the same three-document request)."
537
+ }
538
+ },
539
+ "verify": {
540
+ "workload": "rerank",
541
+ "runtimeConfig": {
542
+ "inference": {
543
+ "rerank": {
544
+ "query": "What API exposes GPU hardware for browser compute workloads?",
545
+ "documents": [
546
+ "WebGPU provides low-level access to graphics processors for compute workloads.",
547
+ "WebSocket provides full-duplex communication over TCP.",
548
+ "CSS grid arranges page layout rows and columns."
549
+ ]
550
+ }
551
+ }
552
+ }
553
+ },
554
+ "vendorBenchmark": null,
555
+ "sourceCheckpointId": "Qwen/Qwen3-Reranker-0.6B",
556
+ "weightPackId": "qwen-3-reranker-0-6b-q4k-ehf16-af32-wp-catalog-v2",
557
+ "manifestVariantId": "qwen-3-reranker-0-6b-q4k-ehf16-af32-mv-exec-v2",
558
+ "artifactCompleteness": "complete",
559
+ "runtimePromotionState": "manifest-owned",
560
+ "weightsRefAllowed": false
561
+ },
562
  {
563
  "modelId": "qwen-3-5-0-8b-q4k-ehaf16",
564
  "family": "qwen3",