{
  "paper": {
    "arxiv_id": "2608.24053",
    "title": "WeMM-Embedding"
  },
  "dataset": {
    "name": "MovieLens-1M paired content/collaborative views",
    "items": 360
  },
  "setup": {
    "seed": 42,
    "alignment_stages": 2,
    "embedding_dimension": 16
  },
  "variants": {
    "stage-1 multimodal alignment": {
      "recall_at_1": 0.1638888888888889,
      "recall_at_10": 0.5611111111111111,
      "mrr": 0.286273656652685
    },
    "WeMM refinement": {
      "recall_at_1": 0.35,
      "recall_at_10": 0.8388888888888889,
      "mrr": 0.5089396375054954
    }
  },
  "matryoshka": {
    "4": {
      "recall_at_1": 0.041666666666666664,
      "recall_at_10": 0.2388888888888889,
      "mrr": 0.11149977910374713
    },
    "8": {
      "recall_at_1": 0.13333333333333333,
      "recall_at_10": 0.5527777777777778,
      "mrr": 0.26739405383847176
    },
    "16": {
      "recall_at_1": 0.35,
      "recall_at_10": 0.8388888888888889,
      "mrr": 0.5089396375054954
    }
  },
  "relative": {
    "recall_at_1_percent": 113.55932203389828,
    "recall_at_10_percent": 49.504950495049506,
    "mrr_percent": 77.78081415397396
  },
  "paper_results": {
    "mmeb_v2_9b": 80.6,
    "online_ab_tests": 14,
    "production_rollout": true
  },
  "scope": "在 MovieLens-1M 公开内容特征与协同共现视图上真实执行两阶段跨模态对齐、细粒度 relevance refinement、cross-scale teacher transfer 和 Matryoshka 截断评测；未加载 WeMM 2B/4B/9B checkpoint、数亿私有样本或微信线上索引。",
  "runtime": {
    "requested_device": "auto",
    "cpu_threads": null,
    "platform": "Darwin arm64",
    "resolved_device": "cpu",
    "torch_version": "2.13.0",
    "accelerator": "arm"
  },
  "seed": 42,
  "schema_version": 2,
  "manifest": {
    "adapter_key": "wemm-embedding",
    "arxiv_id": "2608.24053",
    "title": "WeMM-Embedding: WeChat Multi-Modal Embedding Technical Report",
    "paper_url": "https://arxiv.org/abs/2608.24053",
    "track": "llm",
    "organization": "WeChat Vision, Tencent",
    "published": "2026-08-25",
    "code_url": "https://github.com/Tencent/WeMM-Embedding",
    "topics": [
      "multimodal-foundation-model",
      "embedding",
      "retrieval",
      "matryoshka-representation"
    ],
    "local_code_dir": "src/auto_research/reproductions/wemm_embedding",
    "fidelity": "core_mechanism",
    "evaluation_tier": "l2_public_dataset",
    "datasets": [
      "MovieLens-1M paired content/collaborative views"
    ],
    "baseline": "stage-1 multimodal alignment without refinement",
    "metrics": [
      "Recall@1",
      "Recall@10",
      "MRR"
    ],
    "default_seeds": [
      42
    ],
    "budget": "adapter-defined fixed budget",
    "device_capabilities": [
      "cpu"
    ],
    "online_evidence": [],
    "selection_exception": null,
    "evolve_operators": []
  },
  "provenance": {
    "created_at": "2026-08-26T07:15:31.737360+00:00",
    "code_commit": "98dc66ed800aade784d908e284e3924b16d94bbf",
    "python": "3.12.9",
    "platform": "macOS-26.5.2-arm64-arm-64bit",
    "dataset_dir": "/Users/bytedance/Documents/git_daiwk/auto-research/data",
    "dataset_fingerprint": "e66c261317a7b3179720ef3e3f5f79d7a204b135d1f217e2911b153ad2ce50a0",
    "packages": {
      "auto-research": "0.1.0",
      "numpy": "2.5.2",
      "torch": "2.13.0",
      "transformers": "5.14.1"
    },
    "artifact_path": "docs/reproductions/2608.24053-wemm-embedding/metrics/movielens-1m-seed42.json"
  },
  "evaluation_protocol": {
    "tier": "l2_public_dataset",
    "tier_label": "L2 公开数据集训练",
    "seeds": [
      42
    ],
    "budget": "paper-specific",
    "formal_comparison": false,
    "claim_policy": "single/few-seed smoke result; do not claim a stable improvement"
  },
  "reproduction_fidelity": {
    "level": "core_mechanism",
    "label": "核心机制复现",
    "description": "论文中心算法被实际执行，但生产模型、私有特征或基础设施未复刻。",
    "omitted_core_components": [
      "2B/4B/9B released checkpoints",
      "hundreds of millions of private pairs",
      "WeChat production indexes"
    ]
  },
  "manifest_ref": "reproduction:wemm-embedding"
}