CLIP-ViT-H-14-laion2B-s32B-b79K

Name: CLIP-ViT-H-14-laion2B-s32B-b79K
Author: laion

by laion

OpenCLIP ViT-H/14 trained on LAION-2B: the high-accuracy open CLIP checkpoint for visual search

986Mparams

HuggingFace Run it on your own data, free

Identifiers

Model ID

laion/CLIP-ViT-H-14-laion2B-s32B-b79K

Feature URI

Overview

This is the OpenCLIP ViT-H/14 model trained on the English LAION-2B image-text dataset (32B samples seen). It is one of the most downloaded open CLIP checkpoints and a strong, fully-open alternative to OpenAI CLIP, reaching roughly 78% zero-shot ImageNet top-1: well above the original ViT-L/14.

On Mixpeek, CLIP-ViT-H/14 is a visual embedding extractor: it maps images and video frames into a 1024-dimensional space shared with text, so an agent can search a media library with natural-language queries. Its higher accuracy makes it a good default when retrieval quality matters more than latency.

Architecture

Vision Transformer Huge (ViT-H/14): 32 layers, 1280-dim width, 14x14 patches, paired with a text transformer. Both encoders project into a shared 1024-dim space trained with contrastive (InfoNCE) loss on LAION-2B.

Mixpeek SDK Integration

// No extractor parameter takes a Hugging Face model id (checked against
// GET /v1/discovery/extractors, which returns 13), so CLIP-ViT-H-14-laion2B-s32B-b79K runs
// on your side and the output is upserted through POST
// /v1/namespaces/{namespace_id}/documents/upsert. On Enterprise the other
// path is to upload the weights instead: POST /v1/namespaces/{id}/models
// accepts the huggingface format and a custom plugin loads them.
const res = await fetch(
  "https://api.mixpeek.com/v1/namespaces/ns_your_namespace/documents/upsert",
  {
    method: "POST",
    headers: {
      Authorization: "Bearer API_KEY",
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      collection_id: "col_your_collection",
      documents: [
        {
          document_id: "asset-00412",
          // The vector name has to match a vector index on the collection.
          vectors: { "image-embedding": yourVector },
          payload: { source_key: "archive/2026/asset-00412" },
        },
      ],
    }),
  },
);

// Managed alternative, if this exact model is not the requirement:
// image_extractor@v1 runs google/siglip-base-patch16-224
// (768-d) over a bucket, with no inference of your own.