Publish validated EIDORA schema-v3 ONNX model

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+validation/photo.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,50 @@
+---
+pipeline_tag: feature-extraction
+tags:
+- eidora
+- eidora-model-zoo
+- onnx
+- onnxruntime
+- clip
+- embeddings
+- multimodal
+- vit
+base_model: openai/clip-vit-base-patch32
+---
+
+# EIDORA CLIP ViT-B/32 — Multimodal
+
+Image-and-text model. EIDORA combines the normalized component embeddings using normalized-mean fusion. This combined vector is not the same as CLIP's ordinary image-text similarity score. For text-to-image retrieval, compare the separate Image and Text variants.
+
+Based on `openai/clip-vit-base-patch32`.
+All embeddings are 512-dimensional and L2-normalized.
+
+Sibling variants:
+- CLIP Image: https://huggingface.co/EIDORA/CLIP_ViTB32_Image
+- CLIP Text: https://huggingface.co/EIDORA/CLIP_ViTB32_Text
+- CLIP Multimodal: https://huggingface.co/EIDORA/CLIP_ViTB32_Multimodal
+
+These repositories may not yet be public.
+
+
+## Licensing
+
+The OpenAI CLIP source repository uses MIT.
+The separate redistribution terms of the pretrained
+weights require review before public release.
+
+## Limitations
+
+Similarity does not establish identity, factual correctness,
+authorship or intent. Performance depends on the domain
+and language of the input data.
+
+## EIDORA CLIP Variant Selection
+
+Supports image and text inputs. When both are selected, EIDORA combines their embeddings using normalized-mean fusion. This fused embedding is different from CLIP's usual image-text similarity score. For ordinary text-to-image retrieval, compare the separate Image and Text variants.
+
+Related variants:
+
+- **Image:** https://huggingface.co/EIDORA/CLIP_ViTB32_Image
+- **Text:** https://huggingface.co/EIDORA/CLIP_ViTB32_Text
+- **Multimodal:** https://huggingface.co/EIDORA/CLIP_ViTB32_Multimodal

config.yaml

@@ -0,0 +1,221 @@
+schema_version: 3
+id: EIDORA/CLIP_ViTB32_Multimodal
+name: CLIP ViT-B/32 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: README.md
+  sha256: b1ebd2a22e5c5d8703ae60c0388be5a7f251375a7dcb26641fa4643e5087e218
+  role: asset
+- path: conversion_report.json
+  sha256: 15ad5140537402294eff026ed820e7094625b4be549cf9bf64a9c773ec14edd1
+  role: asset
+- 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
+- path: validation/attention_mask.npy
+  sha256: ae5d428d93c3f5acae10a7ae4944e36da2f5cf5fba57ca047e703b3c0486ec24
+  role: asset
+- path: validation/embedding.npy
+  sha256: 6fe4baa5663769ccab25fcede78b9fd6b73f9d6eb46b4014541fb9eee2538099
+  role: asset
+- path: validation/fused_embedding.npy
+  sha256: 0a787abbc8e338537b146c296d4e23c7759328bd15213388ede319da070e1040
+  role: asset
+- path: validation/image_embedding.npy
+  sha256: 4557f61170cbeadf8b82a6aedc7df3e2f5b70c55c1b02c63f16e26ce1b3224a9
+  role: asset
+- path: validation/input_ids.npy
+  sha256: 0c9c8c962d90e9126314adf19d5fb3b332f1bc1ed8bd2db15f5d3be5d62a86c9
+  role: asset
+- path: validation/photo.png
+  sha256: 1d70a03fc99c16e3aa674dbb9c7a8d0c127b45e1cf416118131451932b2f2389
+  role: asset
+- path: validation/pixels.npy
+  sha256: 49e3049b432634bb2d8c33a78f7a63404e5e9961016b7dd71156cb4a840ce7f3
+  role: asset
+- path: validation/sentence.txt
+  sha256: c0b581bf5fc722a80c9a95b07768db20a9af7edad868de4bfcb7aa759d9e8ab8
+  role: asset
+- path: validation/source_parity.json
+  sha256: 58542124d8c10a0f9e260ceb3a8fcc458e78714676cf350df4d3ffaa8b4d1246
+  role: asset
+- path: validation.yaml
+  sha256: 18ad4d37ccc5bbd5f6150fcee231d62c7f652293288a6e1048c1ad0001a43e2d
+  role: asset
+runtime:
+  execution_providers:
+  - 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:
+        - 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:
+        - 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: cls_projection
+  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

conversion_report.json

@@ -0,0 +1,38 @@
+{
+  "schema_version": 3,
+  "source_framework": "pytorch",
+  "source_model": "openai/clip-vit-base-patch32",
+  "converter": "reuse_independently_verified_onnx_components",
+  "embedding_dimensions": 512,
+  "fusion": "normalized_mean",
+  "components": {
+    "image": {
+      "status": "passed",
+      "max_abs_error": 2.8312206268310547e-07,
+      "sha256": "e7b31f500e645350c2bb0ade2a506e3d9557b2b75338e7941cf4d4db51ee29de",
+      "source_vs_onnx": {
+        "status": "passed",
+        "max_abs_error": 2.8312206268310547e-07,
+        "mean_abs_error": 5.361172839002393e-08,
+        "atol": 0.0001,
+        "rtol": 0.001,
+        "onnxruntime_version": "1.30.0"
+      }
+    },
+    "text": {
+      "status": "passed",
+      "max_abs_error": 2.384185791015625e-07,
+      "sha256": "14e0dd0d24dcf841b6e966d9eb0352586d4aa99238ed840a45580d38bd37bb33",
+      "source_vs_onnx": {
+        "status": "passed",
+        "max_abs_error": 2.384185791015625e-07,
+        "mean_abs_error": 2.0399065547849204e-08,
+        "atol": 0.0001,
+        "rtol": 0.001,
+        "onnxruntime_version": "1.30.0"
+      }
+    }
+  },
+  "shared_space_max_abs_error": 7.450580596923828e-08,
+  "fusion_max_abs_error": 2.1606683731079102e-07
+}

image_encoder.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e7b31f500e645350c2bb0ade2a506e3d9557b2b75338e7941cf4d4db51ee29de
+size 351591848

text_encoder.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:14e0dd0d24dcf841b6e966d9eb0352586d4aa99238ed840a45580d38bd37bb33
+size 253942140

tokenizer/special_tokens_map.json

@@ -0,0 +1,30 @@
+{
+  "bos_token": {
+    "content": "<|startoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer/tokenizer_config.json

@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "49406": {
+      "content": "<|startoftext|>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "49407": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<|startoftext|>",
+  "clean_up_tokenization_spaces": false,
+  "do_lower_case": true,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 77,
+  "pad_token": "<|endoftext|>",
+  "tokenizer_class": "CLIPTokenizer",
+  "unk_token": "<|endoftext|>"
+}

validation-report.json

@@ -0,0 +1,26 @@
+{
+  "desktop_version": "0.2.0",
+  "schema_version": 3,
+  "preparation_version": 2,
+  "providers": {
+    "CPUExecutionProvider": {
+      "status": "passed",
+      "fixture_runs": 23,
+      "scope": "session provider chain; CPU graph-node support allowed",
+      "environment": {
+        "onnxruntime": "1.31.0",
+        "platform": "Linux-7.0.0-38-generic-x86_64-with-glibc2.43",
+        "python": "3.13.16 (main, Oct  9 2026, 15:23:36) [Clang 22.1.3 ]"
+      }
+    }
+  },
+  "environment": {
+    "onnxruntime": "1.31.0",
+    "platform": "Linux-7.0.0-38-generic-x86_64-with-glibc2.43",
+    "python": "3.13.16 (main, Oct  9 2026, 15:23:36) [Clang 22.1.3 ]"
+  },
+  "publication_ready": true,
+  "validation_sha256": "1088394bd33a0a692927508686379b94e3a0e5274fbd497b84b7f87c402d8e6a",
+  "contract_sha256": "65c20b19452a5cedc6874434d8379a280d6c9b4343a4deab6d9772d93ee8d6b0",
+  "package_sha256": "f0efcbecc78a9f65c9e2cf41ed394414fcc5ca2a363dfebf6dca66ced9e79f57"
+}

validation.yaml

@@ -0,0 +1,14 @@
+fixtures:
+- modality: image
+  path: validation/photo.png
+  expected_tensors:
+    pixel_values: validation/pixels.npy
+  expected_vectors: validation/image_embedding.npy
+- modality: text
+  path: validation/sentence.txt
+  expected_tensors:
+    input_ids: validation/input_ids.npy
+    attention_mask: validation/attention_mask.npy
+  expected_vectors: validation/embedding.npy
+note: Original-source image/text references. Normalized-mean fusion reference stored
+  as validation/fused_embedding.npy.

validation/attention_mask.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae5d428d93c3f5acae10a7ae4944e36da2f5cf5fba57ca047e703b3c0486ec24
+size 744

validation/embedding.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6fe4baa5663769ccab25fcede78b9fd6b73f9d6eb46b4014541fb9eee2538099
+size 2176

validation/fused_embedding.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0a787abbc8e338537b146c296d4e23c7759328bd15213388ede319da070e1040
+size 2176

validation/image_embedding.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4557f61170cbeadf8b82a6aedc7df3e2f5b70c55c1b02c63f16e26ce1b3224a9
+size 2176

validation/input_ids.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0c9c8c962d90e9126314adf19d5fb3b332f1bc1ed8bd2db15f5d3be5d62a86c9
+size 744

validation/pixels.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:49e3049b432634bb2d8c33a78f7a63404e5e9961016b7dd71156cb4a840ce7f3
+size 602240

validation/sentence.txt

@@ -0,0 +1 @@
+A photograph of a person.

validation/source_parity.json

@@ -0,0 +1,21 @@
+{
+  "checkpoint": "openai/clip-vit-base-patch32",
+  "certified_image_onnx": "/home/neelesh/eidora-model-zoo-release-20261026/next-batch-20261027/clip-image-release/clip_vit_b32_image_medium_openai_512/model.onnx",
+  "certified_image_sha256": "e7b31f500e645350c2bb0ade2a506e3d9557b2b75338e7941cf4d4db51ee29de",
+  "old_image_onnx": "/home/neelesh/DHSS WORK/EIDORA_ONNX/EIDORA_ONNX/eidora_onnx/dist/CLIP_ViT-B32/image_encoder.onnx",
+  "old_image_sha256": "d93613e6a85cea2b584a00a9dab80337b4d4a00b2a80425fe617bfd46fc186cc",
+  "text_onnx": "/home/neelesh/DHSS WORK/EIDORA_ONNX/EIDORA_ONNX/eidora_onnx/dist/CLIP_ViT-B32/text_encoder.onnx",
+  "text_sha256": "14e0dd0d24dcf841b6e966d9eb0352586d4aa99238ed840a45580d38bd37bb33",
+  "certified_image_max_abs_error": 2.8312206268310547e-07,
+  "old_image_max_abs_error": 0.11951310187578201,
+  "old_vs_certified_image_max_abs_error": 0.11951310932636261,
+  "text_max_abs_error": 2.384185791015625e-07,
+  "corrected_similarity_max_abs_error": 7.450580596923828e-08,
+  "old_similarity_max_abs_error": 0.01277083158493042,
+  "captions": [
+    "A photograph of a person.",
+    "A photograph of buildings.",
+    "A photograph of a landscape."
+  ],
+  "status": "passed"
+}