Upload 2 files
Browse files
openclip/calculate-distances-open.py
CHANGED
@@ -11,30 +11,38 @@ Plan:
|
|
11 |
|
12 |
|
13 |
import sys
|
14 |
-
import torch
|
15 |
-
import open_clip
|
16 |
|
17 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
|
19 |
-
|
|
|
20 |
|
21 |
-
|
22 |
|
23 |
-
mtype=
|
24 |
-
mname=
|
25 |
|
26 |
print("Loading",mtype,mname)
|
27 |
|
|
|
|
|
|
|
|
|
|
|
28 |
cmodel, _, preprocess = open_clip.create_model_and_transforms(mtype,
|
29 |
pretrained=mname)
|
30 |
tokenizer = open_clip.get_tokenizer(mtype)
|
31 |
|
32 |
-
## model = model.to(device)
|
33 |
|
34 |
|
35 |
-
|
36 |
-
|
37 |
-
dictionary=sys.argv[2]
|
38 |
|
39 |
|
40 |
print(f"read in words from {dictionary} now",file=sys.stderr)
|
|
|
11 |
|
12 |
|
13 |
import sys
|
|
|
|
|
14 |
|
15 |
+
if len(sys.argv) <3:
|
16 |
+
print("Need embedding file and a dictionary")
|
17 |
+
print("embedding filename must start with (mtype@stringname). ")
|
18 |
+
exit(1)
|
19 |
+
|
20 |
+
embed_file=sys.argv[1]
|
21 |
+
dictionary=sys.argv[2]
|
22 |
|
23 |
+
dot_index = embed_file.find(".")
|
24 |
+
mstring=embed_file[:dot_index]
|
25 |
|
26 |
+
at_index = mstring.find("@")
|
27 |
|
28 |
+
mtype=mstring[:at_index]
|
29 |
+
mname=mstring[at_index+1:]
|
30 |
|
31 |
print("Loading",mtype,mname)
|
32 |
|
33 |
+
import torch
|
34 |
+
import open_clip
|
35 |
+
from safetensors import safe_open
|
36 |
+
|
37 |
+
|
38 |
cmodel, _, preprocess = open_clip.create_model_and_transforms(mtype,
|
39 |
pretrained=mname)
|
40 |
tokenizer = open_clip.get_tokenizer(mtype)
|
41 |
|
|
|
42 |
|
43 |
|
44 |
+
device=torch.device("cuda")
|
45 |
+
## cmodel.to(device)
|
|
|
46 |
|
47 |
|
48 |
print(f"read in words from {dictionary} now",file=sys.stderr)
|
openclip/generate-embeddings-open.py
CHANGED
@@ -14,7 +14,6 @@ import torch
|
|
14 |
import open_clip
|
15 |
from safetensors.torch import save_file
|
16 |
|
17 |
-
outfile="out.safetensors"
|
18 |
|
19 |
|
20 |
"""
|
@@ -22,14 +21,12 @@ outfile="out.safetensors"
|
|
22 |
You MUST use the same settings when you READ from the output file as well!!
|
23 |
"""
|
24 |
|
25 |
-
#
|
26 |
-
|
27 |
-
#mtype='ViT-
|
28 |
-
#mname='laion2b_s12b_b42k'
|
29 |
-
#mtype='ViT-H-14'
|
30 |
-
#mname='laion2b_s32b_b79k'
|
31 |
mtype='ViT-L-14'
|
32 |
-
mname='
|
|
|
33 |
#### Warning, this requires more than 4GB vram
|
34 |
#mtype='ViT-H-14-quickgelu'
|
35 |
#mname='dfn5b'
|
@@ -38,6 +35,9 @@ mname='laion2b_s32b_b82k'
|
|
38 |
# hf-hub:hf-internal-testing/tiny-open-clip-model'
|
39 |
# for mname
|
40 |
|
|
|
|
|
|
|
41 |
print("Loading",mtype,mname)
|
42 |
|
43 |
cmodel, _, preprocess = open_clip.create_model_and_transforms(
|
|
|
14 |
import open_clip
|
15 |
from safetensors.torch import save_file
|
16 |
|
|
|
17 |
|
18 |
|
19 |
"""
|
|
|
21 |
You MUST use the same settings when you READ from the output file as well!!
|
22 |
"""
|
23 |
|
24 |
+
# See "list_models.txt" for full combination sets
|
25 |
+
|
26 |
+
#mtype='ViT-L-14-336'
|
|
|
|
|
|
|
27 |
mtype='ViT-L-14'
|
28 |
+
mname='openai'
|
29 |
+
|
30 |
#### Warning, this requires more than 4GB vram
|
31 |
#mtype='ViT-H-14-quickgelu'
|
32 |
#mname='dfn5b'
|
|
|
35 |
# hf-hub:hf-internal-testing/tiny-open-clip-model'
|
36 |
# for mname
|
37 |
|
38 |
+
outfile=f"{mtype}-{mname}.safetensors"
|
39 |
+
print("Will save to:")
|
40 |
+
print(" ",outfile)
|
41 |
print("Loading",mtype,mname)
|
42 |
|
43 |
cmodel, _, preprocess = open_clip.create_model_and_transforms(
|