{ "id": "clip-vit-b-16-laion2b-s34b-b88k", "identity": { "family": "openclip", "display_name": "OpenCLIP ViT-B/16 LAION-2B", "hub_id": "laion/CLIP-ViT-B-16-laion2B-s34B-b88K", "api_name": null, "revision": null, "variants": [], "license": "MIT", "commercial_use": true, "commercial_use_note": "MIT weights; LAION notes the datasets are for research, the model license itself has no use restriction.", "release_date": "2023-01", "paper_url": "https://arxiv.org/abs/2210.08402", "repo_url": "https://github.com/mlfoundations/open_clip" }, "task": { "category": "embedding", "modalities_in": [ "image", "text" ], "modalities_out": [ "vector" ], "best_at": "Better than B/32 on fine detail at modest cost.", "worst_at": "English-centric text tower (77 tokens); weak on text in images and counting." }, "io": { "input": { "image": { "resolution": "224x224", "color": "RGB", "normalize": "CLIP mean (0.4815, 0.4578, 0.4082), std (0.2686, 0.2613, 0.2758)" }, "text": { "max_tokens": 77 }, "prompts": "'a photo of a {label}.'; open_clip tokenizer, 77 tokens" }, "output": { "vector": { "dims": 512, "distance": "cosine", "type": "dense" }, "note": "image and text vectors share one space; L2-normalize before cosine" } }, "mixpeek_wiring": { "extractors": [], "feature_uris": [], "vector_indexes": [], "retriever_stages": [ "feature_search", "mmr" ], "composes_with": [ "video frames (VideoFrameExpander) -> image vectors -> feature_search with a text query", "detector boxes (RT-DETR, D-FINE) -> crops -> image vectors per object", "image vectors + E5 text vectors on captions -> feature_search fusion (dbsf)" ] }, "run": { "status": "plugin", "status_reason": "Not served. Would run as a custom extractor plugin (engine/extractors//v1/ with realtime.py for query embedding), loading open_clip_model.safetensors via open_clip hf-hub with trust_remote_code false.", "serve_apps": [], "runtime": "open_clip create_model_and_transforms('hf-hub:')", "weights_format": "safetensors", "trust_remote_code": false, "size_on_disk_mb": 599, "params_millions": 150, "preprocessing": "Resize and center-crop to 224, CLIP mean/std normalize; tokenize to 77.", "postprocessing": "L2-normalize; cosine similarity.", "serving_shape": "both", "batch_size": 32, "serve_batch": null, "scale_to_zero": { "min_replicas": 0, "max_replicas": 1, "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 measurement; plan row 3 measures it in the standalone image." }, "evidence": { "unmeasured": true, "items": [] }, "refs": [ "hf://laion/CLIP-ViT-B-16-laion2B-s34B-b88K@7288da5a0d6f0b51c4a2b27c624837a9236d0112" ], "formats": [ { "ref": "hf://laion/CLIP-ViT-B-16-laion2B-s34B-b88K@7288da5a0d6f0b51c4a2b27c624837a9236d0112", "format": "safetensors", "file": "open_clip_model.safetensors", "loader": "transformers / open_clip from_pretrained; Mixpeek Hub import (api/models/huggingface.py)", "supported_today": true }, { "ref": "hf://laion/CLIP-ViT-B-16-laion2B-s34B-b88K@7288da5a0d6f0b51c4a2b27c624837a9236d0112", "format": "pytorch_bin", "file": "open_clip_pytorch_model.bin", "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": "OpenCLIP ViT-B/16 LAION-2B: embedding model", "description": "OpenCLIP ViT-B/16 LAION-2B is a image and text embedding model (MIT). Can run on Mixpeek as a custom plugin.", "keywords": [ "OpenCLIP ViT-B/16 LAION-2B", "openclip", "embedding", "laion/CLIP-ViT-B-16-laion2B-s34B-b88K", "visual-search", "product-search" ], "faqs": [ { "q": "What is OpenCLIP ViT-B/16 LAION-2B best at?", "a": "Better than B/32 on fine detail at modest cost." }, { "q": "What are the limits of OpenCLIP ViT-B/16 LAION-2B?", "a": "English-centric text tower (77 tokens); weak on text in images and counting." }, { "q": "Can OpenCLIP ViT-B/16 LAION-2B be used commercially?", "a": "Yes. MIT weights; LAION notes the datasets are for research, the model license itself has no use restriction." }, { "q": "How does OpenCLIP ViT-B/16 LAION-2B run on Mixpeek?", "a": "Not served. Would run as a custom extractor plugin (engine/extractors//v1/ with realtime.py for query embedding), loading open_clip_model.safetensors via open_clip hf-hub with trust_remote_code false." }, { "q": "What does OpenCLIP ViT-B/16 LAION-2B output?", "a": "A 512-dimensional dense vector, compared by cosine similarity." } ] }, "last_verified": "2026-11-05", "sources": [ "https://huggingface.co/laion/CLIP-ViT-B-16-laion2B-s34B-b88K", "https://arxiv.org/abs/2210.08402", "https://github.com/mlfoundations/open_clip" ], "registry": null, "templates": [], "usage": { "namespaces": null, "collections": null, "uncounted": true, "templates": 0, "counted_at": null, "source": null }, "facets": { "status": "plugin", "category": "embedding", "family": "openclip", "license": "MIT", "commercial_use": true, "modalities_in": [ "image", "text" ], "modalities_out": [ "vector" ], "use_cases": [ "visual-search", "product-search" ], "recommended": false, "release_date": "2023-01", "params_millions": 150, "size_on_disk_mb": 599, "vector_dims": 512, "vector_distance": "cosine", "weights_format": "safetensors", "formats": [ "pytorch_bin", "safetensors" ], "loadable_today": true, "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": 1 }, "rank_score": 1 }