{
  "id": "mobileclip",
  "identity": {
    "family": "mobileclip",
    "display_name": "MobileCLIP (S0, S1, S2, B)",
    "hub_id": "apple/MobileCLIP-S2",
    "api_name": null,
    "revision": null,
    "variants": [
      "apple/MobileCLIP-S0",
      "apple/MobileCLIP-S1",
      "apple/MobileCLIP-B",
      "apple/MobileCLIP-S2-OpenCLIP",
      "apple/MobileCLIP-B-OpenCLIP"
    ],
    "license": "apple-amlr",
    "commercial_use": false,
    "commercial_use_note": "Apple Machine Learning Research Model license: use only for Research Purposes, which exclude commercial exploitation, product development or use in any commercial product or service.",
    "release_date": "2024-03",
    "paper_url": "https://arxiv.org/abs/2311.17049",
    "repo_url": "https://github.com/apple/ml-mobileclip"
  },
  "task": {
    "category": "embedding",
    "modalities_in": [
      "image",
      "text"
    ],
    "modalities_out": [
      "vector"
    ],
    "best_at": "Very fast image-text embedding on CPU and phones (S0 is the smallest) at CLIP-level zero-shot accuracy.",
    "worst_at": "Cannot be used commercially; lower ceiling than SigLIP so400m."
  },
  "io": {
    "input": {
      "image": {
        "resolution": "256x256",
        "color": "RGB"
      },
      "text": {
        "max_tokens": 77
      },
      "prompts": "CLIP-style captions"
    },
    "output": {
      "vector": {
        "dims": 512,
        "distance": "cosine",
        "type": "dense"
      }
    }
  },
  "mixpeek_wiring": {
    "extractors": [],
    "feature_uris": [],
    "vector_indexes": [],
    "retriever_stages": [],
    "composes_with": [
      "CPU frame embedding at scale to zero, if the license allowed it"
    ]
  },
  "run": {
    "status": "not_runnable",
    "status_reason": "Blocked by license (research only). Technically it would run as a CPU plugin through open_clip (the -OpenCLIP repos ship open_clip_model.safetensors); the mobileclip repos ship .pt pickles.",
    "serve_apps": [],
    "runtime": "open_clip (MobileCLIP-*-OpenCLIP) or the mobileclip package",
    "weights_format": "safetensors",
    "trust_remote_code": false,
    "size_on_disk_mb": 398,
    "params_millions": null,
    "preprocessing": "Resize and center-crop to 256; no mean/std normalization in the reference transform.",
    "postprocessing": "L2-normalize; cosine similarity.",
    "serving_shape": "both",
    "batch_size": 64,
    "serve_batch": null,
    "scale_to_zero": {
      "min_replicas": 0,
      "max_replicas": 1,
      "idle_timeout_s": 600,
      "memory_gb": 1,
      "num_cpus": 1,
      "num_gpus": 0,
      "cold_start_s": {
        "warm_node_s": null,
        "new_node_s": null,
        "source": null
      },
      "cpu_fallback": true
    }
  },
  "cost": {
    "unmeasured": true,
    "measured": null,
    "note": "No committed measurement; plan row 3 measures it in the standalone image."
  },
  "evidence": {
    "unmeasured": true,
    "items": []
  },
  "refs": [
    "hf://apple/MobileCLIP-S2@6da8f9f1c7fa5ae89385ef86bfdee297d105e6f6"
  ],
  "formats": [
    {
      "ref": "hf://apple/MobileCLIP-S2@6da8f9f1c7fa5ae89385ef86bfdee297d105e6f6",
      "format": "pytorch_pt",
      "file": "mobileclip_s2.pt",
      "loader": "torch.load (pickle); Mixpeek refuses pickle weights (shared/security/torch_pickle.py)",
      "supported_today": false
    }
  ],
  "use_cases": [
    "visual-search",
    "product-search"
  ],
  "recommended": {
    "value": false,
    "for": [],
    "reason": null
  },
  "demos": [],
  "seo": {
    "title": "MobileCLIP (S0, S1, S2, B): embedding model",
    "description": "MobileCLIP (S0, S1, S2, B) is a image and text embedding model (apple-amlr). Not runnable on Mixpeek today.",
    "keywords": [
      "MobileCLIP (S0, S1, S2, B)",
      "mobileclip",
      "embedding",
      "apple/MobileCLIP-S2",
      "visual-search",
      "product-search"
    ],
    "faqs": [
      {
        "q": "What is MobileCLIP (S0, S1, S2, B) best at?",
        "a": "Very fast image-text embedding on CPU and phones (S0 is the smallest) at CLIP-level zero-shot accuracy."
      },
      {
        "q": "What are the limits of MobileCLIP (S0, S1, S2, B)?",
        "a": "Cannot be used commercially; lower ceiling than SigLIP so400m."
      },
      {
        "q": "Can MobileCLIP (S0, S1, S2, B) be used commercially?",
        "a": "No. Apple Machine Learning Research Model license: use only for Research Purposes, which exclude commercial exploitation, product development or use in any commercial product or service."
      },
      {
        "q": "How does MobileCLIP (S0, S1, S2, B) run on Mixpeek?",
        "a": "Blocked by license (research only). Technically it would run as a CPU plugin through open_clip (the -OpenCLIP repos ship open_clip_model.safetensors); the mobileclip repos ship .pt pickles."
      },
      {
        "q": "What does MobileCLIP (S0, S1, S2, B) output?",
        "a": "A 512-dimensional dense vector, compared by cosine similarity."
      }
    ]
  },
  "last_verified": "2026-11-05",
  "sources": [
    "https://huggingface.co/apple/MobileCLIP-S2",
    "https://huggingface.co/apple/MobileCLIP-S2-OpenCLIP",
    "https://huggingface.co/apple/MobileCLIP2-S2/blob/main/LICENSE",
    "https://arxiv.org/abs/2311.17049",
    "https://github.com/apple/ml-mobileclip"
  ],
  "registry": null,
  "templates": [],
  "usage": {
    "namespaces": null,
    "collections": null,
    "uncounted": true,
    "templates": 0,
    "counted_at": null,
    "source": null
  },
  "facets": {
    "status": "not_runnable",
    "category": "embedding",
    "family": "mobileclip",
    "license": "apple-amlr",
    "commercial_use": false,
    "modalities_in": [
      "image",
      "text"
    ],
    "modalities_out": [
      "vector"
    ],
    "use_cases": [
      "visual-search",
      "product-search"
    ],
    "recommended": false,
    "release_date": "2024-03",
    "params_millions": null,
    "size_on_disk_mb": 398,
    "vector_dims": 512,
    "vector_distance": "cosine",
    "weights_format": "safetensors",
    "formats": [
      "pytorch_pt"
    ],
    "loadable_today": false,
    "trust_remote_code": false,
    "cpu_fallback": true,
    "num_gpus": 0,
    "memory_gb": 1,
    "min_replicas": 0,
    "serving_shape": "both",
    "served": false,
    "ref_schemes": [
      "hf"
    ],
    "has_evidence": false,
    "cost_measured": false,
    "template_count": 0,
    "usage_namespaces": null,
    "usage_collections": null,
    "usage_uncounted": true,
    "has_demo": false,
    "rank_score": 0
  },
  "rank_score": 0
}