zer0int commited on
Commit
876cbcd
·
verified ·
1 Parent(s): 5e01115

Upload 15 files

Browse files
.gitattributes CHANGED
@@ -58,3 +58,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
 
 
 
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
61
+ restricted_recipes/imagenet_clip_text/manifests/train.jsonl filter=lfs diff=lfs merge=lfs -text
62
+ restricted_recipes/imagenet_handwriting/manifests/train.jsonl filter=lfs diff=lfs merge=lfs -text
HF_PACK_FILE_INVENTORY.tsv ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ path bytes sha256
2
+ HF_PACK_MANIFEST.json 3482 3fadc94515778a58354275b97f208e70f2b5a8567849695fb3b9ad7a1d3e49d5
3
+ LICENSE 1197 d266bde4700b9ad4379b41b24d94b8d152d29cfad40a0c8ede3ac78b1b62437f
4
+ provenance/source_stage/STAGE_FILE_INVENTORY.tsv 607778 f69da88c70f4c83d5eaad5c5a38147de9acebe4e74635e7dee439b6135b9d88a
5
+ provenance/source_stage/STAGE_MANIFEST.json 7036 c5dff83a983349541575e796c0a82a91651e23322c5802550cfa216adb86f3e4
6
+ README.md 8285 eb3f421f4028169545f2fc490bce0e1d82de3fb6d38bc50fe4c127906b73b1b9
7
+ restricted_recipes/imagenet_clip_text/LICENSE 462 7a7ce25d7e8474132e2a0fcc89cfcb3bc90bab35f99bb19ac6996c57069aeda1
8
+ restricted_recipes/imagenet_clip_text/manifests/train.jsonl 66734736 629f59121d1abfdabb4123cf7256ce8bcd40df096fe2419161902882a68fa9de
9
+ restricted_recipes/imagenet_clip_text/manifests/val.jsonl 6661002 8d75bfb7f3a53a2f9057bfccba9141a7cb22b70b44d36ff2cdc9954cf61f9d1c
10
+ restricted_recipes/imagenet_clip_text/NOTICE.md 465 31afb251a242a1d9de7c6be800bc945238a4e312eeacb39f603ab88bcbee2d49
11
+ restricted_recipes/imagenet_handwriting/LICENSE 462 7a7ce25d7e8474132e2a0fcc89cfcb3bc90bab35f99bb19ac6996c57069aeda1
12
+ restricted_recipes/imagenet_handwriting/manifests/train.jsonl 18446263 ef822f28ffc69ded1d307315d94aa986cdd3b6dc88bc017aab913fdac0a3e1d2
13
+ restricted_recipes/imagenet_handwriting/manifests/val.jsonl 1759588 fa7edcb6676ae1c454d944257cb4c2c8c79e262444ec1267973043374824371a
14
+ restricted_recipes/imagenet_handwriting/NOTICE.md 472 c16b7c80434da9576f9b1407ce17412314aa6f377489174e065f8ecbb9ba4205
15
+ runtime_metadata/coco_spright/ATTRIBUTION.md 645 b34d0d61b4b1c622f634fd5becef015a208fbf39c0bc8f1c5df1bc75dfce73b5
16
+ runtime_metadata/coco_spright/coco_spright_reading_supervised_conf999.json 6524793 ab0bf19fa0b4daeae168c989b22deb3bb2b929602dc10c9eed3d800882b713da
17
+ runtime_metadata/coco_spright/LICENSE 383 f42f68b1071bcba99d2d8c995a2fb8c64b1754ab970557b292fbf02a63c89f85
18
+ runtime_metadata/coco_spright/short-coco-spright-train-0_9.json 12406714 72484357d79dfbfb725bb0540550ef25b89979723ead9639c5cca9ff440e3ba6
19
+ runtime_metadata/coco_spright/short-coco-spright-train-0_9_gpt_oss.json 13338254 d88154807e78b5878953d1b99590a37eea29f7d134a89aece25e8859b863287b
20
+ runtime_metadata/coco_spright/short-coco-spright-val-10_11.json 1572559 8fa9e3efa0e79a8b0c344bb50e27980010133c8ce46a5528f56e2ff488bd47fc
21
+ runtime_metadata/coco_spright/short-coco-spright-val-10_11_gpt_oss.json 1712725 c64761203ff5673aabd1cefc1b94ba94b3ae4fe01f6a505d29460ca50302c00b
22
+ runtime_metadata/imagenet/imagenet_wnid_to_class.json 28476 936ac6658228bd78f3778a55fcfcc388a9f21c26200e6cf75af25082dddc3d29
23
+ runtime_metadata/imagenet/LICENSE 366 a87604dee9e46948054670c9344ad6e4c6172d72f7927fb4ee393e8d7f81b6a1
24
+ runtime_metadata/overlay_selection/LICENSE 462 7a7ce25d7e8474132e2a0fcc89cfcb3bc90bab35f99bb19ac6996c57069aeda1
25
+ runtime_metadata/overlay_selection/orthographic_pairs_selected.csv 1487 a32bba0c51fe4138f81ea6ff05ad2d54bf8fd576e2fe0c40ba5ed50043fd9a02
26
+ runtime_metadata/overlay_selection/overlay_words_gmp_diverse.csv 498078 dabd6c9bca20be17df5c378196b6649c568ac9519a75252ed9585ec1d9245f7a
27
+ runtime_metadata/overlay_selection/overlay_words_gmp_diverse.txt 5674 a0cf3b7c0d15564fd0106226657c4b7578a98479a22fe3a948a2143f9a7e7783
28
+ runtime_metadata/textcaps/ATTRIBUTION.md 570 41efc62b3b9e11efe82277e61b1f9127137a5d262c9ac5bd7116fa7e922441a8
29
+ runtime_metadata/textcaps/LICENSE 241 18ed3336ce17e6801e6a44e95b29f299e47e0bc3b47ecce7efdbea6277d762fc
30
+ runtime_metadata/textcaps/manifests/train.jsonl 59847384 4d63bb4d417f08f4f5f6759c9b6b708d11449720b1e6eba2d7ea22a9fd3eb156
31
+ runtime_metadata/textcaps/manifests/validation.jsonl 8872518 069464f70be97cd92a399b68358e9bf34af34c1c9d9c72fedf4cb82a2d18f50b
32
+ shareable_assets/clevr/ATTRIBUTION.md 884 a5e1fb35ca9eda19e0e50a2908aecabf41d781c432d6bc5157ecc9366b722821
33
+ shareable_assets/clevr/clevr_subset_objects_compact.jsonl 6050973 ea9e4bebd8ed5c2ea70c855a969fe0e7aef30e074c2572509389c7f9608c09a1
34
+ shareable_assets/clevr/data/train-00000-of-00002.parquet 201406008 9281641bc51524bd0bd1f963763cab15c6f0aaedaa019a364ad06e34bccbca4e
35
+ shareable_assets/clevr/data/train-00001-of-00002.parquet 129495759 7d77c32e8597ba9be33011241e59652bc1f7383870847fa0e1ebb565e25fb000
36
+ shareable_assets/clevr/data/validation-00000-of-00001.parquet 33911989 9608e43bc9753fcbbf1c762d61d19e4ec768e9d72664a809e8f190277871bc90
37
+ shareable_assets/clevr/LICENSE 241 18ed3336ce17e6801e6a44e95b29f299e47e0bc3b47ecce7efdbea6277d762fc
38
+ shareable_assets/handwriting_overlays/data/train-00000-of-00001.parquet 769590 a78f41daa246dfc392e74e647f4f3dfa500b6f1b575cb884bd23387ca8dfff29
39
+ shareable_assets/handwriting_overlays/LICENSE 462 7a7ce25d7e8474132e2a0fcc89cfcb3bc90bab35f99bb19ac6996c57069aeda1
40
+ shareable_assets/handwriting_overlays/manifest.jsonl 4779940 e9f09283bf687edbaae798eda3e50719fa11f77eedeaa5ecbf228968774f1a2c
41
+ shareable_assets/salt_n_pepper/data/train-00000-of-00001.parquet 43838958 098ed8d54b3f0c5f66d5572cc0f2e757b932117046191e64226587013aea498a
42
+ shareable_assets/salt_n_pepper/LICENSE 462 7a7ce25d7e8474132e2a0fcc89cfcb3bc90bab35f99bb19ac6996c57069aeda1
43
+ shareable_assets/salt_n_pepper/NOTICE.md 333 cf7923515b7639293c9899834c99999eabde548fb9d9d1c589c2e24cefeaf025
44
+ THIRD_PARTY_NOTICES.md 565 1e19ffd7985fccc2e9706ada088c46cdb73ceb7d8a40826ede592a97f0f85562
HF_PACK_MANIFEST.json ADDED
@@ -0,0 +1,112 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "schema_version": 1,
3
+ "source_stage_manifest_sha256": "c5dff83a983349541575e796c0a82a91651e23322c5802550cfa216adb86f3e4",
4
+ "source_stage_inventory_sha256": "f69da88c70f4c83d5eaad5c5a38147de9acebe4e74635e7dee439b6135b9d88a",
5
+ "target_shard_mib": 192,
6
+ "parquet_batch_size": 100,
7
+ "configs": {
8
+ "clevr": {
9
+ "train": 3824,
10
+ "validation": 394
11
+ },
12
+ "handwriting_overlays": {
13
+ "train": 206
14
+ },
15
+ "salt_n_pepper": {
16
+ "train": 186
17
+ }
18
+ },
19
+ "shards": [
20
+ {
21
+ "config": "clevr",
22
+ "split": "train",
23
+ "path": "shareable_assets/clevr/data/train-00000-of-00002.parquet",
24
+ "rows": 2329,
25
+ "parquet_bytes": 201406008,
26
+ "sha256": "9281641bc51524bd0bd1f963763cab15c6f0aaedaa019a364ad06e34bccbca4e",
27
+ "embedded_image_bytes": 201263278
28
+ },
29
+ {
30
+ "config": "clevr",
31
+ "split": "train",
32
+ "path": "shareable_assets/clevr/data/train-00001-of-00002.parquet",
33
+ "rows": 1495,
34
+ "parquet_bytes": 129495759,
35
+ "sha256": "7d77c32e8597ba9be33011241e59652bc1f7383870847fa0e1ebb565e25fb000",
36
+ "embedded_image_bytes": 129404494
37
+ },
38
+ {
39
+ "config": "clevr",
40
+ "split": "validation",
41
+ "path": "shareable_assets/clevr/data/validation-00000-of-00001.parquet",
42
+ "rows": 394,
43
+ "parquet_bytes": 33911989,
44
+ "sha256": "9608e43bc9753fcbbf1c762d61d19e4ec768e9d72664a809e8f190277871bc90",
45
+ "embedded_image_bytes": 33887394
46
+ },
47
+ {
48
+ "config": "handwriting_overlays",
49
+ "split": "train",
50
+ "path": "shareable_assets/handwriting_overlays/data/train-00000-of-00001.parquet",
51
+ "rows": 206,
52
+ "parquet_bytes": 769590,
53
+ "sha256": "a78f41daa246dfc392e74e647f4f3dfa500b6f1b575cb884bd23387ca8dfff29",
54
+ "embedded_image_bytes": 717034
55
+ },
56
+ {
57
+ "config": "salt_n_pepper",
58
+ "split": "train",
59
+ "path": "shareable_assets/salt_n_pepper/data/train-00000-of-00001.parquet",
60
+ "rows": 186,
61
+ "parquet_bytes": 43838958,
62
+ "sha256": "098ed8d54b3f0c5f66d5572cc0f2e757b932117046191e64226587013aea498a",
63
+ "embedded_image_bytes": 43825879
64
+ }
65
+ ],
66
+ "copied_raw_files": 36,
67
+ "copied_raw_bytes": 209864650,
68
+ "verification": {
69
+ "stage": {
70
+ "ok": true,
71
+ "expected_image_counts": {
72
+ "shareable_assets/clevr/images/train": 3824,
73
+ "shareable_assets/clevr/images/validation": 394,
74
+ "shareable_assets/handwriting_overlays/images": 206,
75
+ "shareable_assets/salt_n_pepper": 186
76
+ },
77
+ "observed_image_counts": {
78
+ "shareable_assets/clevr/images/train": 3824,
79
+ "shareable_assets/clevr/images/validation": 394,
80
+ "shareable_assets/handwriting_overlays/images": 206,
81
+ "shareable_assets/salt_n_pepper": 186
82
+ },
83
+ "metadata_files_checked_for_absolute_paths": 14,
84
+ "absolute_path_hits": 0,
85
+ "denied_artifacts": 0
86
+ },
87
+ "output": {
88
+ "ok": true,
89
+ "loose_image_files": 0,
90
+ "forbidden_artifacts": 0,
91
+ "absolute_path_hits": 0,
92
+ "parquet_rows": {
93
+ "clevr/train": 3824,
94
+ "clevr/validation": 394,
95
+ "handwriting_overlays/train": 206,
96
+ "salt_n_pepper/train": 186
97
+ }
98
+ },
99
+ "output_final": {
100
+ "ok": true,
101
+ "loose_image_files": 0,
102
+ "forbidden_artifacts": 0,
103
+ "absolute_path_hits": 0,
104
+ "parquet_rows": {
105
+ "clevr/train": 3824,
106
+ "clevr/validation": 394,
107
+ "handwriting_overlays/train": 206,
108
+ "salt_n_pepper/train": 186
109
+ }
110
+ }
111
+ }
112
+ }
LICENSE ADDED
@@ -0,0 +1,16 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ CLIP-MUX TRAINING DATA PACK — MIXED COMPONENT LICENSES
2
+
3
+ This repository does not have a single uniform license. The applicable terms are determined by component path.
4
+
5
+ 1. Project-authored material explicitly marked CC0 1.0 Universal is dedicated/waived under CC0 to the extent the contributor holds the relevant rights.
6
+ Canonical terms: https://creativecommons.org/publicdomain/zero/1.0/
7
+
8
+ 2. The CLEVR derivative and TextCaps-derived runtime metadata are distributed under CC BY 4.0 and require the attribution described in their component directories.
9
+ Canonical terms: https://creativecommons.org/licenses/by/4.0/
10
+
11
+ 3. SPRIGHT-COCO-derived labels/metadata are distributed subject to the Intel Research Use License used by the upstream SPRIGHT-COCO dataset.
12
+ Upstream license: https://huggingface.co/datasets/SPRIGHT-T2I/spright_coco/blob/main/LICENSE
13
+
14
+ 4. ImageNet images are not distributed. ImageNet/WordNet-derived metadata and any locally constructed ImageNet derivatives remain subject to applicable upstream terms. Project-authored recipe contributions are separately marked CC0 where appropriate.
15
+
16
+ See each component's LICENSE / NOTICE / ATTRIBUTION file for scope and provenance.
README.md ADDED
@@ -0,0 +1,148 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ pretty_name: CLIP-MUX Training Data Pack
3
+ language:
4
+ - en
5
+ license: other
6
+ license_name: mixed-component-licenses
7
+ license_link: https://huggingface.co/datasets/zer0int/CLIP-Cross-Attn-MUX-Training-Data-Pack/blob/main/LICENSE
8
+ tags:
9
+ - clip
10
+ - vision-language
11
+ - image-text
12
+ - robustness
13
+ - training-data
14
+ - reproducibility
15
+ configs:
16
+ - config_name: clevr
17
+ data_files:
18
+ - split: train
19
+ path: shareable_assets/clevr/data/train-*.parquet
20
+ - split: validation
21
+ path: shareable_assets/clevr/data/validation-*.parquet
22
+ - config_name: handwriting_overlays
23
+ data_files:
24
+ - split: train
25
+ path: shareable_assets/handwriting_overlays/data/train-*.parquet
26
+ - config_name: salt_n_pepper
27
+ data_files:
28
+ - split: train
29
+ path: shareable_assets/salt_n_pepper/data/train-*.parquet
30
+ ---
31
+
32
+ # CLIP-MUX Training Data Pack
33
+
34
+ This repository is the data/metadata companion for reproducing the CLIP-MUX / x-attention CLIP training pipeline. It is **not one monolithic dataset under one license**. Each component is independently scoped and carries its own `LICENSE` or `NOTICE` file.
35
+
36
+ The pack intentionally separates:
37
+
38
+ - assets that can be redistributed directly;
39
+ - runtime labels/manifests derived from upstream datasets;
40
+ - recipes for datasets that users must construct from separately obtained restricted inputs.
41
+
42
+ ## Contents and licenses
43
+
44
+ | Path | Contents | Distribution / license |
45
+ | --- | --- | --- |
46
+ | `shareable_assets/clevr/` | Selected/preprocessed CLEVR subset used by CLIP-MUX | **CC BY 4.0**, inherited from CLEVR; see component `LICENSE` and `ATTRIBUTION.md` |
47
+ | `shareable_assets/handwriting_overlays/` | Handwriting overlay assets created for this project | **CC0 1.0 Universal** for rights held by the contributor |
48
+ | `shareable_assets/salt_n_pepper/` | Salt/pepper image set created for this project | **CC0 1.0 Universal** for rights held by the contributor |
49
+ | `runtime_metadata/overlay_selection/` | Overlay word bank and orthographic grouping metadata | **CC0 1.0 Universal** |
50
+ | `runtime_metadata/textcaps/` | CLIP-MUX runtime manifests derived from TextCaps | **CC BY 4.0**; see component `LICENSE` and `ATTRIBUTION.md` |
51
+ | `runtime_metadata/coco_spright/` | CLIP-MUX short/reworded SPRIGHT-COCO labels and trusted-reading metadata | **Intel Research Use License**, following the upstream SPRIGHT-COCO material from which these labels are derived |
52
+ | `runtime_metadata/imagenet/` | Convenience ImageNet/WordNet class metadata | **Upstream terms apply**; no independent CC0 grant is asserted for upstream metadata |
53
+ | `restricted_recipes/imagenet_clip_text/` | Portable recipe/manifests for building the digital-text ImageNet derivative | Project-authored recipe contributions are **CC0 1.0**; this does **not** grant rights to ImageNet images or other upstream material |
54
+ | `restricted_recipes/imagenet_handwriting/` | Portable recipe/manifests for building the handwriting ImageNet derivative | Project-authored recipe contributions are **CC0 1.0**; this does **not** grant rights to ImageNet images or other upstream material |
55
+
56
+ The repository-level license metadata is therefore `other`: **the applicable license is determined by component path**.
57
+
58
+ ## Redistributed assets
59
+
60
+ ### CLEVR subset
61
+
62
+ The CLEVR component is a selected/preprocessed derivative of the CLEVR dataset. The original CLEVR data is released under CC BY 4.0.
63
+
64
+ Changes made for this training corpus include square-crop/selection preprocessing intended to preserve depicted objects for CLIP training; samples for which an acceptable object-preserving crop could not be produced were excluded. Compact object metadata used by the training pipeline is included alongside the images.
65
+
66
+ Upstream project: <https://cs.stanford.edu/people/jcjohns/clevr/>
67
+
68
+ ### Handwriting overlays
69
+
70
+ These overlay assets were created for this project and are released under CC0 1.0 Universal to the extent the contributor holds copyright or related rights in them.
71
+
72
+ ### Salt and pepper image set
73
+
74
+ This project-specific image set is released under CC0 1.0 Universal to the extent the contributor holds copyright or related rights in the material. Some images may have been produced using generative-AI tools. CC0 applies only to rights, if any, held by the contributor; it does not create rights in material that is not protectable or grant third-party trademark, privacy, publicity, or other rights.
75
+
76
+ ## Upstream datasets reconstructed by the preparation script
77
+
78
+ ### TextCaps
79
+
80
+ CLIP-MUX uses TextCaps images and human reference captions/OCR information. The development copy was obtained through `shwetkm/TextCaps-Caption-Summary`; generated summary/QA fields are not part of the CLIP-MUX runtime representation. The preparation tool downloads the upstream Hugging Face dataset and materializes the image/manifests layout expected by the training loader.
81
+
82
+ TextCaps is distributed under CC BY 4.0.
83
+
84
+ - Official TextCaps site: <https://textvqa.org/textcaps/download/>
85
+ - Acquisition source used by this project: <https://huggingface.co/datasets/shwetkm/TextCaps-Caption-Summary>
86
+
87
+ ### SPRIGHT-COCO
88
+
89
+ The preparation tool obtains images from `SPRIGHT-T2I/spright_coco` and reconstructs the `data/<shard>/<key>.jpg` layout expected by the training code. This pack contains the CLIP-MUX training label variants and trusted-reading metadata, but does not duplicate the upstream image payload.
90
+
91
+ SPRIGHT-COCO is distributed under the Intel Research Use License.
92
+
93
+ Upstream dataset: <https://huggingface.co/datasets/SPRIGHT-T2I/spright_coco>
94
+
95
+ ## Restricted-input recipes: ImageNet
96
+
97
+ ImageNet images are **not distributed** in this repository. Users must obtain ImageNet independently under its applicable terms and provide a local ImageNet root to the preparation script.
98
+
99
+ The recipe manifests in `restricted_recipes/` contain portable relative references and project-authored construction metadata for reproducing the digital-text and handwriting-overlay derivatives used during training. Any CC0 dedication in those folders covers only project-authored recipe contributions and does not change the terms governing ImageNet or WordNet material.
100
+
101
+ ImageNet access/terms: <https://image-net.org/>
102
+
103
+ ## Intended use
104
+
105
+ This pack exists to make the accompanying CLIP-MUX training code reproducible without requiring users to reconstruct project-specific preprocessing, label shortening, word-avoidance rewrites, overlay selections, or small shareable corpora by hand.
106
+
107
+ It is designed for research and model-training workflows subject to the license of each component and any independently obtained upstream datasets.
108
+
109
+ ## Data preparation
110
+
111
+ The Git repository provides a preparation utility that will eventually support a workflow of the form:
112
+
113
+ ```bash
114
+ python prepare_training_data.py \
115
+ --data-root /path/to/clip_mux_data \
116
+ --imagenet-root /path/to/ILSVRC2012
117
+ ```
118
+
119
+ The utility is responsible for downloading shareable project assets, obtaining TextCaps/SPRIGHT from their upstream locations, validating runtime metadata, and constructing ImageNet-derived datasets locally from the user's own ImageNet copy.
120
+
121
+ ## Attribution and provenance
122
+
123
+ Attribution required by upstream licenses is recorded in the relevant component directories. The component-scoped license/notice files take precedence over any general description in this card.
124
+
125
+ ## No warranty
126
+
127
+ The data and metadata are provided for reproducibility and research use without warranty. Users are responsible for complying with the terms of upstream datasets and for determining whether their intended use is permitted.
128
+
129
+ ## Hugging Face representation
130
+
131
+ The redistributable image payloads are stored as Parquet with embedded image bytes. The
132
+ repository exposes three explicit Hugging Face dataset configurations:
133
+
134
+ ```python
135
+ from datasets import load_dataset
136
+
137
+ clevr = load_dataset("REPO_ID", "clevr")
138
+ handwriting = load_dataset("REPO_ID", "handwriting_overlays")
139
+ salt_pepper = load_dataset("REPO_ID", "salt_n_pepper")
140
+ ```
141
+
142
+ The remaining runtime metadata and reconstruction recipes intentionally stay as ordinary
143
+ JSON/JSONL/CSV/TXT repository files. They are consumed by the CLIP-MUX data-preparation
144
+ utility rather than treated as independent tabular dataset configurations.
145
+
146
+ Parquet image records embed the original bytes and store only portable basenames in the
147
+ image `path` field. Local staging paths are not part of the published representation.
148
+
THIRD_PARTY_NOTICES.md ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ # Third-party notices
2
+
3
+ This repository combines project-authored material with components derived from separately licensed datasets.
4
+
5
+ - **CLEVR** — CC BY 4.0: <https://cs.stanford.edu/people/jcjohns/clevr/>
6
+ - **TextCaps** — CC BY 4.0: <https://textvqa.org/textcaps/download/>
7
+ - **SPRIGHT-COCO** — Intel Research Use License: <https://huggingface.co/datasets/SPRIGHT-T2I/spright_coco>
8
+ - **ImageNet** — images are not redistributed; users obtain ImageNet independently: <https://image-net.org/>
9
+
10
+ See the component directories for exact scope and attribution.
provenance/source_stage/STAGE_FILE_INVENTORY.tsv ADDED
The diff for this file is too large to render. See raw diff
 
provenance/source_stage/STAGE_MANIFEST.json ADDED
@@ -0,0 +1,260 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "schema_version": 2,
3
+ "purpose": "CLIP-MUX authoritative training-data HF staging capture",
4
+ "policy": {
5
+ "whitelist_only": true,
6
+ "copies_imagenet_pixels": false,
7
+ "copies_textcaps_pixels": false,
8
+ "copies_spright_pixels": false,
9
+ "copies_checkpoints": false,
10
+ "copies_debug_or_harmony_traces": false,
11
+ "absolute_machine_paths_allowed_in_path_fields": false
12
+ },
13
+ "collections": [
14
+ {
15
+ "name": "coco_train_json",
16
+ "destination": "runtime_metadata/coco_spright/short-coco-spright-train-0_9.json",
17
+ "files": 1,
18
+ "bytes": 12406714,
19
+ "sanitized_strings": 0,
20
+ "notes": []
21
+ },
22
+ {
23
+ "name": "coco_train_gpt_json",
24
+ "destination": "runtime_metadata/coco_spright/short-coco-spright-train-0_9_gpt_oss.json",
25
+ "files": 1,
26
+ "bytes": 13338254,
27
+ "sanitized_strings": 0,
28
+ "notes": []
29
+ },
30
+ {
31
+ "name": "coco_val_json",
32
+ "destination": "runtime_metadata/coco_spright/short-coco-spright-val-10_11.json",
33
+ "files": 1,
34
+ "bytes": 1572559,
35
+ "sanitized_strings": 0,
36
+ "notes": []
37
+ },
38
+ {
39
+ "name": "coco_val_gpt_json",
40
+ "destination": "runtime_metadata/coco_spright/short-coco-spright-val-10_11_gpt_oss.json",
41
+ "files": 1,
42
+ "bytes": 1712725,
43
+ "sanitized_strings": 0,
44
+ "notes": []
45
+ },
46
+ {
47
+ "name": "coco_train_reading_json",
48
+ "destination": "runtime_metadata/coco_spright/coco_spright_reading_supervised_conf999.json",
49
+ "files": 1,
50
+ "bytes": 6524793,
51
+ "sanitized_strings": 0,
52
+ "notes": [
53
+ "runtime-pruned rows=781"
54
+ ]
55
+ },
56
+ {
57
+ "name": "imagenet_wnid_json",
58
+ "destination": "runtime_metadata/imagenet/imagenet_wnid_to_class.json",
59
+ "files": 1,
60
+ "bytes": 28476,
61
+ "sanitized_strings": 0,
62
+ "notes": []
63
+ },
64
+ {
65
+ "name": "overlay_selection/overlay_words_gmp_diverse.csv",
66
+ "destination": "runtime_metadata/overlay_selection/overlay_words_gmp_diverse.csv",
67
+ "files": 1,
68
+ "bytes": 498078,
69
+ "sanitized_strings": 0,
70
+ "notes": []
71
+ },
72
+ {
73
+ "name": "overlay_selection/overlay_words_gmp_diverse.txt",
74
+ "destination": "runtime_metadata/overlay_selection/overlay_words_gmp_diverse.txt",
75
+ "files": 1,
76
+ "bytes": 5674,
77
+ "sanitized_strings": 0,
78
+ "notes": []
79
+ },
80
+ {
81
+ "name": "overlay_selection/orthographic_pairs_selected.csv",
82
+ "destination": "runtime_metadata/overlay_selection/orthographic_pairs_selected.csv",
83
+ "files": 1,
84
+ "bytes": 1487,
85
+ "sanitized_strings": 0,
86
+ "notes": []
87
+ },
88
+ {
89
+ "name": "textcaps_train_manifest",
90
+ "destination": "runtime_metadata/textcaps/manifests/train.jsonl",
91
+ "files": 1,
92
+ "bytes": 59847384,
93
+ "sanitized_strings": 0,
94
+ "notes": [
95
+ "runtime-pruned rows=21953"
96
+ ]
97
+ },
98
+ {
99
+ "name": "textcaps_validation_manifest",
100
+ "destination": "runtime_metadata/textcaps/manifests/validation.jsonl",
101
+ "files": 1,
102
+ "bytes": 8872518,
103
+ "sanitized_strings": 0,
104
+ "notes": [
105
+ "runtime-pruned rows=3166"
106
+ ]
107
+ },
108
+ {
109
+ "name": "imagenet_clip_text_train_manifest",
110
+ "destination": "restricted_recipes/imagenet_clip_text/manifests/train.jsonl",
111
+ "files": 1,
112
+ "bytes": 66734736,
113
+ "sanitized_strings": 60000,
114
+ "notes": [
115
+ "rows=60000"
116
+ ]
117
+ },
118
+ {
119
+ "name": "imagenet_handwriting_train_manifest",
120
+ "destination": "restricted_recipes/imagenet_handwriting/manifests/train.jsonl",
121
+ "files": 1,
122
+ "bytes": 18446263,
123
+ "sanitized_strings": 27514,
124
+ "notes": [
125
+ "rows=13757"
126
+ ]
127
+ },
128
+ {
129
+ "name": "imagenet_clip_text_val_manifest",
130
+ "destination": "restricted_recipes/imagenet_clip_text/manifests/val.jsonl",
131
+ "files": 1,
132
+ "bytes": 6661002,
133
+ "sanitized_strings": 6000,
134
+ "notes": [
135
+ "rows=6000"
136
+ ]
137
+ },
138
+ {
139
+ "name": "imagenet_handwriting_val_manifest",
140
+ "destination": "restricted_recipes/imagenet_handwriting/manifests/val.jsonl",
141
+ "files": 1,
142
+ "bytes": 1759588,
143
+ "sanitized_strings": 2614,
144
+ "notes": [
145
+ "rows=1307"
146
+ ]
147
+ },
148
+ {
149
+ "name": "clevr_train_images",
150
+ "destination": "shareable_assets/clevr/images/train",
151
+ "files": 3824,
152
+ "bytes": 330667772,
153
+ "sanitized_strings": 0,
154
+ "notes": [
155
+ "PNG-only; count=3824"
156
+ ]
157
+ },
158
+ {
159
+ "name": "clevr_validation_images",
160
+ "destination": "shareable_assets/clevr/images/validation",
161
+ "files": 394,
162
+ "bytes": 33887394,
163
+ "sanitized_strings": 0,
164
+ "notes": [
165
+ "PNG-only; count=394"
166
+ ]
167
+ },
168
+ {
169
+ "name": "clevr_metadata_jsonl",
170
+ "destination": "shareable_assets/clevr/clevr_subset_objects_compact.jsonl",
171
+ "files": 1,
172
+ "bytes": 6050973,
173
+ "sanitized_strings": 4218,
174
+ "notes": [
175
+ "rows=4218"
176
+ ]
177
+ },
178
+ {
179
+ "name": "handwriting_overlay_manifest",
180
+ "destination": "shareable_assets/handwriting_overlays/manifest.jsonl",
181
+ "files": 1,
182
+ "bytes": 4779940,
183
+ "sanitized_strings": 412,
184
+ "notes": [
185
+ "rows=206"
186
+ ]
187
+ },
188
+ {
189
+ "name": "handwriting_overlay_images",
190
+ "destination": "shareable_assets/handwriting_overlays/images",
191
+ "files": 206,
192
+ "bytes": 717034,
193
+ "sanitized_strings": 0,
194
+ "notes": [
195
+ "PNG-only; count=206"
196
+ ]
197
+ },
198
+ {
199
+ "name": "salt_n_pepper_images",
200
+ "destination": "shareable_assets/salt_n_pepper",
201
+ "files": 186,
202
+ "bytes": 43825879,
203
+ "sanitized_strings": 0,
204
+ "notes": [
205
+ "PNG-only; count=186"
206
+ ]
207
+ }
208
+ ],
209
+ "verification": {
210
+ "ok": true,
211
+ "absolute_path_hits": 0,
212
+ "denied_artifacts": 0,
213
+ "unexpected_gpt_oss_artifacts": 0,
214
+ "restricted_pixel_leaks": 0,
215
+ "hard_image_count_checks": {
216
+ "shareable_assets/clevr/images/train": {
217
+ "expected": 3824,
218
+ "observed": 3824,
219
+ "ok": true
220
+ },
221
+ "shareable_assets/clevr/images/validation": {
222
+ "expected": 394,
223
+ "observed": 394,
224
+ "ok": true
225
+ },
226
+ "shareable_assets/handwriting_overlays/images": {
227
+ "expected": 206,
228
+ "observed": 206,
229
+ "ok": true
230
+ },
231
+ "shareable_assets/salt_n_pepper": {
232
+ "expected": 186,
233
+ "observed": 186,
234
+ "ok": true
235
+ }
236
+ },
237
+ "hard_jsonl_row_checks": {
238
+ "runtime_metadata/textcaps/manifests/train.jsonl": {
239
+ "expected": 21953,
240
+ "observed": 21953,
241
+ "ok": true
242
+ },
243
+ "runtime_metadata/textcaps/manifests/validation.jsonl": {
244
+ "expected": 3166,
245
+ "observed": 3166,
246
+ "ok": true
247
+ },
248
+ "restricted_recipes/imagenet_clip_text/manifests/train.jsonl": {
249
+ "expected": 60000,
250
+ "observed": 60000,
251
+ "ok": true
252
+ },
253
+ "restricted_recipes/imagenet_clip_text/manifests/val.jsonl": {
254
+ "expected": 6000,
255
+ "observed": 6000,
256
+ "ok": true
257
+ }
258
+ }
259
+ }
260
+ }
restricted_recipes/imagenet_clip_text/LICENSE ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ CC0 1.0 UNIVERSAL
2
+
3
+ This component is released under CC0 1.0 Universal to the extent the contributor holds Copyright and Related Rights in the material.
4
+
5
+ Canonical deed and legal terms:
6
+ https://creativecommons.org/publicdomain/zero/1.0/
7
+
8
+ CC0 does not grant patent or trademark rights, does not override third-party rights, and does not create copyright where none exists. No attribution is required by the contributor for material covered by this CC0 dedication.
restricted_recipes/imagenet_clip_text/NOTICE.md ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ # ImageNet digital-text recipe
2
+
3
+ The CC0 file in this directory applies only to project-authored recipe/manifest contributions to the extent the contributor holds rights in them.
4
+
5
+ **It does not license or redistribute ImageNet images.** Users must obtain ImageNet independently and comply with its applicable terms. Relative source-image references in these manifests are reconstruction pointers, not redistributed source images.
6
+
7
+ ImageNet: <https://image-net.org/>
restricted_recipes/imagenet_clip_text/manifests/train.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:629f59121d1abfdabb4123cf7256ce8bcd40df096fe2419161902882a68fa9de
3
+ size 66734736
restricted_recipes/imagenet_clip_text/manifests/val.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
restricted_recipes/imagenet_handwriting/LICENSE ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ CC0 1.0 UNIVERSAL
2
+
3
+ This component is released under CC0 1.0 Universal to the extent the contributor holds Copyright and Related Rights in the material.
4
+
5
+ Canonical deed and legal terms:
6
+ https://creativecommons.org/publicdomain/zero/1.0/
7
+
8
+ CC0 does not grant patent or trademark rights, does not override third-party rights, and does not create copyright where none exists. No attribution is required by the contributor for material covered by this CC0 dedication.
restricted_recipes/imagenet_handwriting/NOTICE.md ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ # ImageNet handwriting-overlay recipe
2
+
3
+ The CC0 file in this directory applies only to project-authored recipe/manifest contributions to the extent the contributor holds rights in them.
4
+
5
+ **It does not license or redistribute ImageNet images.** Users must obtain ImageNet independently and comply with its applicable terms. Relative source-image references in these manifests are reconstruction pointers, not redistributed source images.
6
+
7
+ ImageNet: <https://image-net.org/>
restricted_recipes/imagenet_handwriting/manifests/train.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ef822f28ffc69ded1d307315d94aa986cdd3b6dc88bc017aab913fdac0a3e1d2
3
+ size 18446263
restricted_recipes/imagenet_handwriting/manifests/val.jsonl ADDED
The diff for this file is too large to render. See raw diff