{
  "id": "clap-htsat-tiny",
  "identity": {
    "family": "laion-clap",
    "display_name": "LAION CLAP HTSAT-tiny (630k-audioset-best)",
    "hub_id": "lukewys/laion_clap",
    "api_name": null,
    "revision": "630k-audioset-best.pt",
    "variants": [],
    "license": "CC0-1.0",
    "commercial_use": true,
    "commercial_use_note": "The checkpoint repo is CC0-1.0 (public domain dedication). Training data includes AudioSet and LAION-Audio-630K; no license restriction on the weights.",
    "release_date": "2023-02",
    "paper_url": "https://arxiv.org/abs/2211.06687",
    "repo_url": "https://github.com/LAION-AI/CLAP"
  },
  "task": {
    "category": "embedding",
    "modalities_in": [
      "audio",
      "text"
    ],
    "modalities_out": [
      "vector"
    ],
    "best_at": "Sound-alike search over short audio windows (jingles, music beds, effects) and text-to-audio search (\"upbeat jingle with bells\") in one 512-d space.",
    "worst_at": "Speech content (use whisper + E5); exact duplicate detection where an acoustic fingerprint algorithm is cheaper; audio longer than the 5 s window is seen only as overlapping windows."
  },
  "io": {
    "input": {
      "audio": {
        "sample_rate": 48000,
        "window_s": 5,
        "hop_s": 2.5,
        "channels": "mono"
      },
      "text": "free-text description of a sound"
    },
    "output": {
      "vector": {
        "dims": 512,
        "distance": "cosine",
        "type": "dense"
      }
    }
  },
  "mixpeek_wiring": {
    "extractors": [
      {
        "extractor": "audio_fingerprint_extractor@v1",
        "selected_by": "fixed"
      }
    ],
    "feature_uris": [
      "mixpeek://audio_fingerprint_extractor@v1/laion__clap_htsat_tiny"
    ],
    "vector_indexes": [
      "audio_fingerprint_extractor_v1_embedding"
    ],
    "retriever_stages": [
      "feature_search"
    ],
    "composes_with": [
      "audio or video -> 5 s windows with 2.5 s hop -> CLAP vector per window -> feature_search by audio clip or text"
    ]
  },
  "run": {
    "status": "served",
    "status_reason": "Ray Serve app in the inference registry (query embeds); audio_fingerprint_extractor loads the same checkpoint in-process (AudioFingerprintBatchService).",
    "serve_apps": [
      "laion__clap_htsat_tiny"
    ],
    "runtime": "laion_clap.CLAP_Module(enable_fusion=False) + load_ckpt (HTSAT-tiny audio encoder, RoBERTa text encoder)",
    "weights_format": "pickle",
    "trust_remote_code": false,
    "size_on_disk_mb": 1864,
    "params_millions": null,
    "preprocessing": "Extract audio from video, resample to 48 kHz, slice 5 s windows every 2.5 s; text goes through the CLAP text tower.",
    "postprocessing": "get_audio_embedding_from_filelist / get_text_embedding (512 dims), normalized when normalize_embeddings is true.",
    "serving_shape": "both",
    "batch_size": 8,
    "serve_batch": null,
    "scale_to_zero": {
      "min_replicas": 0,
      "max_replicas": 4,
      "idle_timeout_s": 600,
      "memory_gb": 2,
      "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 throughput or cost measurement for this model."
  },
  "evidence": {
    "unmeasured": true,
    "items": []
  },
  "refs": [
    "hf://lukewys/laion_clap@630k-audioset-best.pt"
  ],
  "formats": [
    {
      "ref": "hf://lukewys/laion_clap@630k-audioset-best.pt",
      "format": "pytorch_pt",
      "file": "7 files, e.g. 630k-audioset-best.pt",
      "loader": "torch.load (pickle); Mixpeek refuses pickle weights (shared/security/torch_pickle.py)",
      "supported_today": false
    }
  ],
  "use_cases": [
    "audio-search"
  ],
  "recommended": {
    "value": false,
    "for": [],
    "reason": null
  },
  "demos": [],
  "seo": {
    "title": "LAION CLAP HTSAT-tiny (630k-audioset-best): embedding model",
    "description": "LAION CLAP HTSAT-tiny (630k-audioset-best) is a audio and text embedding model (CC0-1.0). Served on Mixpeek, scaling to zero when idle.",
    "keywords": [
      "LAION CLAP HTSAT-tiny (630k-audioset-best)",
      "laion-clap",
      "embedding",
      "lukewys/laion_clap",
      "audio-search"
    ],
    "faqs": [
      {
        "q": "What is LAION CLAP HTSAT-tiny (630k-audioset-best) best at?",
        "a": "Sound-alike search over short audio windows (jingles, music beds, effects) and text-to-audio search (\"upbeat jingle with bells\") in one 512-d space."
      },
      {
        "q": "What are the limits of LAION CLAP HTSAT-tiny (630k-audioset-best)?",
        "a": "Speech content (use whisper + E5); exact duplicate detection where an acoustic fingerprint algorithm is cheaper; audio longer than the 5 s window is seen only as overlapping windows."
      },
      {
        "q": "Can LAION CLAP HTSAT-tiny (630k-audioset-best) be used commercially?",
        "a": "Yes. The checkpoint repo is CC0-1.0 (public domain dedication). Training data includes AudioSet and LAION-Audio-630K; no license restriction on the weights."
      },
      {
        "q": "How does LAION CLAP HTSAT-tiny (630k-audioset-best) run on Mixpeek?",
        "a": "Ray Serve app in the inference registry (query embeds); audio_fingerprint_extractor loads the same checkpoint in-process (AudioFingerprintBatchService)."
      }
    ]
  },
  "last_verified": "2026-11-05",
  "sources": [
    "https://huggingface.co/lukewys/laion_clap",
    "https://huggingface.co/api/models/lukewys/laion_clap",
    "https://arxiv.org/abs/2211.06687",
    "https://github.com/LAION-AI/CLAP",
    "server/engine/inference/laion/clap_htsat_tiny/services.py",
    "server/engine/extractors/audio_fingerprint_extractor/v1/definition.py",
    "server/engine/extractors/audio_fingerprint_extractor/v1/processors/processors.py"
  ],
  "registry": {
    "laion__clap_htsat_tiny": {
      "deployment": "laion__clap_htsat_tiny",
      "service_id": "laion/clap-htsat-tiny",
      "routes": "engine/inference/laion/clap_htsat_tiny/routes.py",
      "route_prefix": "/audio",
      "enabled": {
        "expr": "importlib.util.find_spec('laion_clap') is not None"
      },
      "description": "CLAP HTSAT-tiny audio embedding service. Generates 512-d L2-normalized audio embeddings for audio fingerprint matching and similarity search. Also embeds TEXT queries into the same space (contrastive text-audio model) for natural-language soundtrack search.",
      "capabilities": {
        "category": "embedding",
        "modality_input": [
          "audio",
          "text"
        ],
        "modality_output": "vector",
        "supports_batch": true,
        "reranking_mode": null
      },
      "vector_config": {
        "dimensions": 512,
        "distance": "cosine",
        "vector_type": "dense"
      },
      "serve": {
        "min_replicas": 0,
        "max_replicas": 4,
        "idle_timeout_s": 600,
        "target_ongoing_requests": 8,
        "num_cpus": 1,
        "num_gpus": 0,
        "memory_gb": 2,
        "pinnable": false,
        "placement": "worker",
        "enabled": true
      },
      "interactive_twin": false,
      "interactive_serve": null,
      "schemas": {
        "input": "engine.inference.laion.clap_htsat_tiny.models.AudioEmbedRequest",
        "output": "engine.inference.laion.clap_htsat_tiny.models.AudioEmbedResponse",
        "config": null
      }
    }
  },
  "templates": [],
  "usage": {
    "namespaces": null,
    "collections": null,
    "uncounted": true,
    "templates": 0,
    "counted_at": null,
    "source": null
  },
  "facets": {
    "status": "served",
    "category": "embedding",
    "family": "laion-clap",
    "license": "CC0-1.0",
    "commercial_use": true,
    "modalities_in": [
      "audio",
      "text"
    ],
    "modalities_out": [
      "vector"
    ],
    "use_cases": [
      "audio-search"
    ],
    "recommended": false,
    "release_date": "2023-02",
    "params_millions": null,
    "size_on_disk_mb": 1864,
    "vector_dims": 512,
    "vector_distance": "cosine",
    "weights_format": "pickle",
    "formats": [
      "pytorch_pt"
    ],
    "loadable_today": false,
    "trust_remote_code": false,
    "cpu_fallback": true,
    "num_gpus": 0,
    "memory_gb": 2,
    "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": 3
  },
  "rank_score": 3
}