schema_version: 3
id: EIDORA/CLIP_ViTB32_Multimodal
name: CLIP_ViTB32_Multimodal
version: 0.1.0
modality: multimodal
license: other
bibtex: "@inproceedings{radford2021learning,\n  title={Learning Transferable Visual\
  \ Models From Natural Language Supervision},\n  author={Radford, Alec and Kim, Jong\
  \ Wook and Hallacy, Chris and Ramesh, Aditya and Goh, Gabriel and Agarwal, Sandhini\
  \ and Sastry, Girish and Askell, Amanda and Mishkin, Pamela and Clark, Jack and\
  \ Krueger, Gretchen and Sutskever, Ilya},\n  booktitle={ICML},\n  year={2021}\n}"
compatibility:
  min_eidora_version: 0.2.0
artifacts:
- path: image_encoder.onnx
  sha256: e7b31f500e645350c2bb0ade2a506e3d9557b2b75338e7941cf4d4db51ee29de
  role: graph
- path: text_encoder.onnx
  sha256: 14e0dd0d24dcf841b6e966d9eb0352586d4aa99238ed840a45580d38bd37bb33
  role: graph
- path: tokenizer/merges.txt
  sha256: 9fd691f7c8039210e0fced15865466c65820d09b63988b0174bfe25de299051a
  role: asset
- path: tokenizer/special_tokens_map.json
  sha256: 2cdb3b8331a60c92fc1e55a13e9fd61fd2293c5a51275fdcccd62b780052530e
  role: asset
- path: tokenizer/tokenizer.json
  sha256: 8e019f9349caf5f9bb92514b0e4fc00c92841aef6ae8df9d833d975c623b7c72
  role: asset
- path: tokenizer/tokenizer_config.json
  sha256: 39bae5313c2da6467866dee758e754d69f0e302a875027b142531cb9c51c7c28
  role: asset
- path: tokenizer/vocab.json
  sha256: 5047b556ce86ccaf6aa22b3ffccfc52d391ea4accdab9c2f2407da5b742d4363
  role: asset
runtime:
  execution_providers:
  - CUDAExecutionProvider
  - CoreMLExecutionProvider
  - CPUExecutionProvider
  fusion: normalized_mean
  components:
    image:
      runtime:
        graph: image_encoder.onnx
        inputs:
        - name: pixel_values
          source: pixels
          dtype: float32
          shape:
          - batch
          - 3
          - 224
          - 224
        output:
          name: embedding
          dtype: float32
          shape:
          - batch
          - 512
        execution_providers:
        - CUDAExecutionProvider
        - CoreMLExecutionProvider
        - CPUExecutionProvider
      preparation:
        width: 224
        height: 224
        layout: NCHW
        resize: short_edge_crop
        resize_rounding: truncate
        interpolation: bicubic
        rescale: 0.00392156862745098
        mean:
        - 0.48145466
        - 0.4578275
        - 0.40821073
        std:
        - 0.26862954
        - 0.26130258
        - 0.27577711
        alpha_background:
        - 0
        - 0
        - 0
        short_edge: 224
        parameters:
          resize:
            default: short_edge_crop
            choices:
            - stretch
            - fill_crop
            - fit_pad
            - short_edge_crop
          interpolation:
            default: bicubic
            choices:
            - nearest
            - bilinear
            - bicubic
            - lanczos
          crop_position:
            default: center
          short_edge:
            default: 224
            minimum: 224
            maximum: 2048
    text:
      runtime:
        graph: text_encoder.onnx
        inputs:
        - name: input_ids
          source: token_ids
          dtype: int64
          shape:
          - batch
          - 77
        - name: attention_mask
          source: attention_mask
          dtype: int64
          shape:
          - batch
          - 77
        output:
          name: embedding
          dtype: float32
          shape:
          - batch
          - 512
        execution_providers:
        - CUDAExecutionProvider
        - CoreMLExecutionProvider
        - CPUExecutionProvider
      preparation:
        tokenizer_path: tokenizer
        max_length: 77
        padding_side: right
        truncation_side: right
        add_special_tokens: true
        prefix: ''
        suffix: ''
        parameters:
          truncation_side:
            default: right
embedding:
  dimensions: 512
  normalized: true
  similarity: cosine
  shared_space:
    id: openai/clip-vit-base-patch32
    revision: b97b0100e55e367c057773c2a614676470b0d575
metadata:
  summary: CLIP ViT-B/32 image and text encoders combined using EIDORA normalized-mean
    fusion.
  trained_on: Publicly available image-caption data collected from the internet; the
    exact training dataset was not released.
  feature_type: embedding
  pooling: image CLS projection and text end-of-text projection; EIDORA normalized-mean
    fusion
  group:
    id: multimodal
    label: Multimodal Models
    order: 10
  provenance:
    base_model: openai/clip-vit-base-patch32
    source_repository: https://huggingface.co/openai/clip-vit-base-patch32
    original_name: CLIP ViT-B/32
    original_model_url: https://github.com/openai/CLIP
    paper_title: Learning Transferable Visual Models From Natural Language Supervision
    paper_url: https://arxiv.org/abs/2103.00020
    training_data: Publicly available image-caption data collected from the internet;
      the exact training dataset was not released.
    upstream_license: OpenAI CLIP source repository is MIT. A separate explicit pretrained-weight
      redistribution statement was not found during review, so public redistribution
      remains blocked pending review.
    authors:
    - Alec Radford
    - Jong Wook Kim
    - Chris Hallacy
    - Aditya Ramesh
    - Gabriel Goh
    - Sandhini Agarwal
    - Girish Sastry
    - Amanda Askell
    - Pamela Mishkin
    - Jack Clark
    - Gretchen Krueger
    - Ilya Sutskever
    source_model: openai/clip-vit-base-patch32
    source_revision: b97b0100e55e367c057773c2a614676470b0d575
    exported_opset: 17
    export_date: '2026-10-26'
    exporter_version: eidora-onnx 1.0.0