Datasets:
key stringlengths 9 9 | caption stringlengths 1 1.63k | features list |
|---|---|---|
000000001 | This is the best recipe I have ever tried for Cuban bread. I lived in Key West... Cuban Recipes, Bread Recipes, Cooking Recipes, Cuban Desserts, Pan Cubano Recipe, Cuban Bread, Cuban Sandwich, Sandwiches, Recipe From Scratch | [
-1.111328125,
-0.4658203125,
-1.0283203125,
-0.8671875,
-0.250732421875,
-0.0149688720703125,
0.65869140625,
0.27685546875,
1.4326171875,
0.08172607421875,
-0.434326171875,
-0.2213134765625,
-0.1856689453125,
0.22802734375,
-0.01081085205078125,
-0.1624755859375,
-0.06634521484375,
... |
000000003 | Red Square is a city square in Moscow, Russia. | [
0.4609375,
0.049468994140625,
1.541015625,
-0.41259765625,
-0.384033203125,
-0.2130126953125,
-0.36865234375,
0.4365234375,
-0.249755859375,
-0.0284271240234375,
0.47998046875,
0.1859130859375,
0.10943603515625,
0.505859375,
0.2423095703125,
-1.1572265625,
-0.2091064453125,
0.10266... |
000000004 | Photography lenses are on the table. | [
-0.375,
1.0654296875,
-0.243896484375,
0.90869140625,
0.7236328125,
0.21630859375,
-0.20947265625,
0.302734375,
-0.2347412109375,
0.01898193359375,
0.537109375,
0.1202392578125,
0.428466796875,
-0.25146484375,
-0.0234832763671875,
0.2391357421875,
-0.7294921875,
0.10699462890625,
... |
000000006 | Woman exercising by skipping a rope | [
0.6044921875,
-0.392333984375,
2.064453125,
0.6328125,
0.40625,
0.095947265625,
-1.189453125,
0.7734375,
0.86328125,
0.140380859375,
0.01549530029296875,
-0.322509765625,
0.300537109375,
-0.375,
-0.54345703125,
0.51220703125,
0.042510986328125,
-0.044464111328125,
0.0861206054687... |
000000007 | Shoe's the Word Art Print | [
0.82421875,
0.5068359375,
-0.55615234375,
0.1165771484375,
-0.38134765625,
-0.1099853515625,
-0.79248046875,
0.487548828125,
-0.460693359375,
-0.0377197265625,
-0.35693359375,
0.1336669921875,
0.8349609375,
-0.2149658203125,
-0.309326171875,
-0.44189453125,
0.286865234375,
-0.11621... |
000000008 | 7 personal finance tips every small business owner needs | [
0.1326904296875,
-0.419677734375,
0.767578125,
0.281005859375,
-0.5439453125,
0.199951171875,
-0.6123046875,
0.127197265625,
-0.00659942626953125,
0.439453125,
-0.205322265625,
0.1790771484375,
-0.35107421875,
0.260986328125,
-0.310302734375,
0.463134765625,
-0.199951171875,
0.5893... |
000000009 | A bathroom at Staybridge Suites Toronto - Vaughan South | [
0.798828125,
0.11798095703125,
0.58154296875,
-1.2822265625,
0.144287109375,
0.0272369384765625,
0.81494140625,
-0.39111328125,
0.476806640625,
0.468505859375,
0.89453125,
-0.162841796875,
-0.309326171875,
0.28369140625,
0.54443359375,
0.2027587890625,
0.478515625,
0.6416015625,
... |
000000010 | Last Stop for the 7 Train | by <PERSON>. | [
0.54638671875,
0.062255859375,
0.81103515625,
-0.57177734375,
0.01220703125,
-0.297607421875,
-0.3076171875,
0.147705078125,
0.133544921875,
-1.095703125,
0.84033203125,
0.416015625,
-0.043212890625,
0.242431640625,
-0.427734375,
-0.49267578125,
-0.10369873046875,
-0.301513671875,
... |
000000011 | Dumbbell, barbell and workout in the gym. Copy space on black grunge wall and board background. Dumbbell, barbell, weights and workout in the gym. Copy space on stock photos | [
0.74560546875,
-1.3232421875,
-0.97705078125,
0.7333984375,
-0.147216796875,
-0.0090179443359375,
0.398681640625,
0.65478515625,
-0.640625,
-0.05224609375,
-0.1396484375,
0.2159423828125,
-0.1343994140625,
-0.08538818359375,
-0.2880859375,
-0.00998687744140625,
0.318359375,
-0.1844... |
000000012 | Looking for high-quality carbon fiber parts ? We have all the parts you need to customize your exotic car. | [
0.7822265625,
0.11370849609375,
-1.3349609375,
0.59375,
-0.1588134765625,
-0.1611328125,
-0.84228515625,
0.26708984375,
0.01543426513671875,
0.152587890625,
-0.48291015625,
-0.466064453125,
0.1431884765625,
0.326171875,
0.55859375,
-0.09619140625,
0.650390625,
-0.304931640625,
0.... |
000000014 | From all of us at Heritage Iron, we wish you a very Merry Christmas! Photo by Super T. IH Farmall Gold Demos owned by <PERSON> of Sebring, FL. International Tractors, Very Merry Christmas, Puns, <PERSON>, Gold, Merry Christmas, Clean Puns, Word Games | [
0.6552734375,
0.1951904296875,
-0.205810546875,
-0.5224609375,
0.1336669921875,
-0.1334228515625,
1.0107421875,
-0.1763916015625,
-0.189208984375,
0.2200927734375,
0.74267578125,
0.2435302734375,
-0.1240234375,
-0.177734375,
0.1365966796875,
-0.95166015625,
0.089599609375,
0.169921... |
000000015 | One of the completed artworks. | [
1.396484375,
0.00327301025390625,
-1.4326171875,
1.6484375,
0.458251953125,
0.451416015625,
0.2264404296875,
-0.6474609375,
-0.5361328125,
-0.231201171875,
-0.020660400390625,
-0.99072265625,
-0.45361328125,
-0.256591796875,
-0.951171875,
0.00411224365234375,
0.74365234375,
1.23046... |
000000017 | The seal was spotted playing in the shallows. Photo / <PERSON> | [
-0.61669921875,
0.57275390625,
-0.46484375,
-1.5546875,
0.68115234375,
0.64990234375,
-0.5458984375,
0.167236328125,
-1.22265625,
-0.32666015625,
-0.423095703125,
0.27294921875,
0.249755859375,
0.91552734375,
0.8486328125,
-0.242919921875,
0.25634765625,
-0.12017822265625,
-0.136... |
000000018 | The Endura Back Pack uses padded shoulder straps for added comfort | [
0.1092529296875,
-0.296142578125,
0.455810546875,
-0.5419921875,
0.03900146484375,
0.0496826171875,
0.369873046875,
-0.83203125,
0.258056640625,
-0.239501953125,
0.2269287109375,
0.00572967529296875,
-0.462646484375,
-0.720703125,
0.33837890625,
0.176513671875,
0.288330078125,
-0.1... |
000000019 | Lighting Catalog The Home Depot 2018 Home Depot, Led, Interior Lighting, Your Space, Pendant Lighting, Cabinet, Storage, Modern, Table | [
0.2064208984375,
-0.1961669921875,
-0.488525390625,
-0.70166015625,
-0.08837890625,
-0.27783203125,
-1.115234375,
-0.31640625,
-1.0068359375,
-0.59912109375,
0.42529296875,
0.16259765625,
-0.033599853515625,
-0.1151123046875,
0.0264434814453125,
1,
-0.1380615234375,
-0.022491455078... |
000000020 | Overseas beauty raises 10 years ``pearl clam'', after cruel heart cut, discovers the value of great pearls! | [
0.26318359375,
-0.1900634765625,
-1.81640625,
-1.5419921875,
-0.08148193359375,
0.018402099609375,
0.88916015625,
0.4228515625,
0.3828125,
0.063720703125,
-0.6279296875,
-0.406005859375,
0.037689208984375,
-0.058746337890625,
0.00102996826171875,
-0.81640625,
0.74853515625,
-0.2939... |
000000021 | Local fishermen usually sells Tilapia to tourists and trekkers staying at the Lake. | [
-0.269775390625,
-0.37646484375,
0.054656982421875,
-0.05169677734375,
-0.465576171875,
-0.0005130767822265625,
0.07110595703125,
0.416259765625,
0.060211181640625,
-0.0285491943359375,
-0.16748046875,
-0.01299285888671875,
0.25830078125,
-0.109619140625,
0.322021484375,
-0.5986328125,... |
000000022 | Multiple Teddy Bears iPhone Case for all iPhones - Choose your Phone | [
0.012298583984375,
0.3037109375,
-0.302734375,
0.27490234375,
-0.1461181640625,
0.1417236328125,
-0.0731201171875,
0.55322265625,
0.395263671875,
-0.498779296875,
-0.70654296875,
0.5341796875,
0.6865234375,
0.3876953125,
0.016815185546875,
0.430419921875,
0.74072265625,
0.043121337... |
000000024 | A bartender making a cocktail at Roosevelt Room | [
0.330078125,
0.17724609375,
-0.1883544921875,
-0.7109375,
0.080078125,
-0.3203125,
-0.11920166015625,
0.2177734375,
0.7451171875,
-0.9033203125,
-0.08843994140625,
-0.39794921875,
-0.39208984375,
-0.1810302734375,
-0.0885009765625,
-0.255126953125,
-0.00007164478302001953,
-0.18933... |
000000025 | Kids Sky blue color craft. Kids blue color background with shapes of the deer and stars. The Abstract children illustration with a variety light sky blue color stock illustration | [
-0.57470703125,
-0.61328125,
-1.2490234375,
-0.9892578125,
-0.1005859375,
-0.195068359375,
-0.736328125,
-0.354248046875,
1.072265625,
-0.10003662109375,
0.1224365234375,
-0.68212890625,
-0.142578125,
-0.2393798828125,
0.431640625,
-0.276123046875,
0.9345703125,
-0.5224609375,
0.... |
000000026 | Bring together tart lemon with sweet coconut in these nutrition-packed Lemon Coconut Chia Energy Balls. These portable snacks help you curb your hunger when you need that 3:00 PM pick-me-up! | [
-0.2388916015625,
-0.084228515625,
-0.237060546875,
-1.029296875,
-0.277587890625,
0.419677734375,
0.953125,
-0.224365234375,
0.37060546875,
-0.302734375,
0.3134765625,
-0.303955078125,
0.06646728515625,
-0.435791015625,
0.227294921875,
-0.28955078125,
-0.75048828125,
0.03170776367... |
000000027 | A Day To Remember ``Bad Vibrations'' Logo Graphic T-Shirt | [
1.146484375,
-0.26904296875,
-0.434326171875,
0.7578125,
0.59765625,
-0.5205078125,
0.64208984375,
0.0286865234375,
-0.4921875,
0.2135009765625,
0.796875,
-0.359619140625,
0.0860595703125,
0.70068359375,
-0.10919189453125,
-0.69775390625,
-0.204833984375,
-0.393310546875,
0.20019... |
000000030 | An illustration of a man in silhouette hands and arms raised stretching up to the sky in praise or joy | [
0.026519775390625,
-0.89501953125,
-1.7890625,
0.435791015625,
0.01544952392578125,
-0.189697265625,
-0.0242919921875,
0.1529541015625,
0.5771484375,
0.162353515625,
0.0927734375,
-0.060638427734375,
-0.333984375,
-0.0071868896484375,
-0.2939453125,
0.69580078125,
0.2890625,
0.1875... |
000000031 | The Circa Building Final Update | [
0.6435546875,
0.08612060546875,
1.78125,
1.052734375,
-0.177001953125,
-0.64013671875,
-1.2529296875,
-0.074951171875,
-0.048858642578125,
0.1910400390625,
-0.047454833984375,
-0.1851806640625,
-0.271728515625,
-0.220947265625,
-0.302490234375,
-0.165283203125,
0.13330078125,
-0.18... |
000000032 | Builder <PERSON> Construction Worker Trowel Tool. A cartoon builder or bricklayer construction worker holding a masons brick laying trowel hand tool and giving royalty free illustration | [
-0.541015625,
-0.130126953125,
-0.86865234375,
-0.376953125,
0.329833984375,
-0.30126953125,
-0.049468994140625,
-0.0780029296875,
-0.5263671875,
0.25732421875,
0.7470703125,
0.00693511962890625,
-0.478271484375,
0.60595703125,
0.013458251953125,
-0.7236328125,
0.60888671875,
-0.18... |
bulk-cc12m-features — teacher-tower image features for CC12M
Precomputed image-tower features for 10,968,539 CC12M images (all 2,176 shards of pixparse/cc12m-wds), from two teachers: CLIP ViT-B/16 (LAION-2B) (laion/CLIP-ViT-B-16-laion2B-s34B-b88K, 512-d projection space, with captions) and SigLIP base-patch16-384 (google/siglip-base-patch16-384, 768-d, see the tower section for its distinct preprocessing). Built for CLIP distillation research — a student can train against these targets with zero teacher inference, cutting distillation compute by roughly a third. Captions are included per sample, so contrastive and text-side objectives work from this file set alone.
~30 GB total across both towers (vs 1.18 TB for the source
images): fp16, one .pt file per source shard per tower.
Exact preprocessing — CLIP tower (the part banks usually leave undocumented)
Feature parity requires bit-level preprocessing agreement. Everything below was verified against live-tower recomputation at cosine 1.00000 before extraction started, and the extractor re-proved it at startup on reference images:
- Model:
laion/CLIP-ViT-B-16-laion2B-s34B-b88K(open_clip weights, deterministically converted to transformersCLIPModelformat) - Activation: plain GELU — NOT QuickGELU. (The open_clip default for some checkpoints silently differs; the mismatch reads as cosine ~0.975 and is easy to miss.)
- Geometry:
torchvision.transforms.Resize(224, InterpolationMode.BICUBIC)(shortest side) →CenterCrop(224)on the PIL image - Normalization: CLIP mean
(0.48145466, 0.4578275, 0.40821073), std(0.26862954, 0.26130258, 0.27577711) - Compute: fp32, TF32 disabled; readout
get_image_features(CLS → visual projection) - Storage: fp16, UNNORMALIZED (L2-normalize at load if you need unit vectors; mean norm ≈ 12.76)
Layout
clip_b16_laion2b/features_0000.pt ... features_2175.pt # one per wds shard
clip_b16_laion2b/train-00000-of-00064.parquet ... # parquet (viewer/datasets)
siglip_b16_384/features_0000.pt ... features_2175.pt # 2nd tower, 768-d
siglip_b16_384/train-00000-of-00064.parquet ... # parquet (viewer/datasets)
siglip_b16_384/ledger.jsonl # its per-shard counts
ledger.jsonl # per-shard counts
extract.log # full run log
Each tower ships in TWO forms with identical content: compact fp16 .pt
shards (2,176, matching the wds shard numbering) and consolidated
parquet (64 files, columns key, caption (CLIP config only),
features as fp32 lists carrying fp16 precision — the same convention as
the companion COCO bank) for the dataset viewer and datasets loading:
from datasets import load_dataset
ds = load_dataset("AbstractPhil/bulk-cc12m-features",
"clip_b16_laion2b", split="train", streaming=True)
row = next(iter(ds)) # {'key', 'caption', 'features': [512 floats]}
Each .pt (loadable with weights_only=True) contains:
| field | type | meaning |
|---|---|---|
keys |
list[str] |
sample keys, exactly the pixparse/cc12m-wds keys |
captions |
list[str] |
the paired CC12M captions |
emb |
float16 (N, 512) |
unnormalized projection features, row-aligned |
tower / precision / normalized |
str/str/bool | provenance markers |
Shards hold 5,040–5,041 samples; row order within a shard is decode-completion order (keys are the join handle, not position).
Second tower: siglip_b16_384/ (added 2026-07-29)
SigLIP base-patch16-384 (google/siglip-base-patch16-384)
image features for the same 10,968,539 images: 768-d, fp16, unnormalized,
same shard numbering and keys. Preprocessing differs from the CLIP tower and
is the SigLIP AutoProcessor default: 384×384 warp resize (bicubic, no
crop), mean/std 0.5. Extraction was verified before spend by two gates: the
worker path vs the reference processor (cos 1.000000) and reproduction of
this tower's stored COCO features from the companion bank (cos 0.999996+).
Captions are not duplicated here — join them from clip_b16_laion2b/ by key.
Fields per shard: keys, emb (N, 768), provenance markers (no captions).
Usage
import torch
import torch.nn.functional as F
from huggingface_hub import hf_hub_download
p = hf_hub_download("AbstractPhil/bulk-cc12m-features",
"clip_b16_laion2b/features_0000.pt", repo_type="dataset")
d = torch.load(p, map_location="cpu", weights_only=True)
z = F.normalize(d["emb"].float(), dim=-1) # (5041, 512) unit vectors
print(d["keys"][0], d["captions"][0])
To pair features with pixels, stream the matching tar from
pixparse/cc12m-wds
and join on keys (shard numbering is identical).
Provenance and quality
- Extracted 2026-07-27 in a single 8.1-hour streaming pass (download → decode → embed → discard images), ~376 img/s sustained on one consumer GPU.
- Zero decode errors and zero download failures across all 2,176 shards
(
ledger.jsonlhas per-shard counts; truncated-JPEG tolerance was enabled). - Full-bank integrity sweep after extraction: every file loads, key/caption/ embedding row counts agree with the ledger, no non-finite values, feature norms stable (12.75–12.77) across the whole run.
- Startup parity gate: 8 held-out reference images through the extraction path vs independently stored features of the same tower — cosine 1.00000.
Licensing note
These are derived features and captions, not images. CC12M imagery remains the property of its owners; captions and the underlying URL list are provided by Google's Conceptual 12M under its stated terms, and the image snapshot mirrored by pixparse/cc12m-wds. Intended for research use.
References
- Changpinyo et al., Conceptual 12M: Pushing Web-Scale Image-Text Pre-Training To Recognize Long-Tail Visual Concepts — https://arxiv.org/abs/2102.08981
- Radford et al., Learning Transferable Visual Models From Natural Language Supervision (CLIP) — https://arxiv.org/abs/2103.00020
- Cherti et al., Reproducible scaling laws for contrastive language-image learning (OpenCLIP / LAION-2B) — https://arxiv.org/abs/2212.07143
- Companion COCO-2017 bank (34 towers): AbstractPhil/bulk-coco-features
- First consumer of this bank: AbstractPhil/clip-vitb-mini-distilled
- Downloads last month
- 42