NimaBoscarino
commited on
Commit
•
67613a8
1
Parent(s):
bd62bca
Create laion2b-en-vit_embeddings.py
Browse files- laion2b-en-vit_embeddings.py +56 -0
laion2b-en-vit_embeddings.py
ADDED
@@ -0,0 +1,56 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import numpy as np
|
2 |
+
import datasets
|
3 |
+
|
4 |
+
from huggingface_hub import HfApi
|
5 |
+
api = HfApi()
|
6 |
+
repo_files = list(api.dataset_info(repo_id="laion/laion2b-en-vit-h-14-embeddings").siblings)
|
7 |
+
filenames = [x.rfilename for x in repo_files]
|
8 |
+
img_embs = [x for x in filenames if x.startswith("img_emb/")]
|
9 |
+
|
10 |
+
|
11 |
+
class LAIONEmbeddingsConfig(datasets.BuilderConfig):
|
12 |
+
def __init__(self, **kwargs):
|
13 |
+
super(LAIONEmbeddingsConfig, self).__init__(version=datasets.Version("1.0.0"), **kwargs)
|
14 |
+
|
15 |
+
|
16 |
+
class LAIONEmbeddings(datasets.GeneratorBasedBuilder):
|
17 |
+
VERSION = datasets.Version("1.0.0")
|
18 |
+
|
19 |
+
BUILDER_CONFIGS = [
|
20 |
+
LAIONEmbeddingsConfig()
|
21 |
+
]
|
22 |
+
|
23 |
+
def _get_features(self) -> datasets.Features:
|
24 |
+
return datasets.Features({
|
25 |
+
"embedding": datasets.Sequence(datasets.Value("float32")),
|
26 |
+
})
|
27 |
+
|
28 |
+
def _info(self):
|
29 |
+
features = self._get_features()
|
30 |
+
|
31 |
+
return datasets.DatasetInfo(
|
32 |
+
features=features,
|
33 |
+
)
|
34 |
+
|
35 |
+
def _split_generators(self, dl_manager):
|
36 |
+
main_url = "https://huggingface.co/datasets/laion/laion2b-en-vit-h-14-embeddings/resolve/main/"
|
37 |
+
archive_paths = dl_manager.download([main_url + x for x in img_embs])
|
38 |
+
|
39 |
+
return [
|
40 |
+
datasets.SplitGenerator(
|
41 |
+
name=datasets.Split.TRAIN,
|
42 |
+
gen_kwargs={
|
43 |
+
"chunks": archive_paths,
|
44 |
+
"split": "train",
|
45 |
+
},
|
46 |
+
),
|
47 |
+
]
|
48 |
+
|
49 |
+
def _generate_examples(self, chunks, split):
|
50 |
+
for chunk in chunks:
|
51 |
+
file = np.DataSource().open(chunk)
|
52 |
+
data = np.load(file.name)
|
53 |
+
for example in data:
|
54 |
+
yield "", {
|
55 |
+
"embedding": example
|
56 |
+
}
|