Upload 2 files
Browse files- README.md +11 -3
- graph-textmodels.py +113 -0
README.md
CHANGED
@@ -9,14 +9,22 @@ Primary tools are:
|
|
9 |
* "graph-embeddings.py": plots graph of full values of two embeddings
|
10 |
|
11 |
|
12 |
-
## calculate-distances.py
|
13 |
|
14 |
Loads the generated embeddings, reads in a word, calculates "distance" to every
|
15 |
embedding, and then shows the closest "neighbours".
|
16 |
|
17 |
-
To run this requires the files "embeddings.safetensors" and "dictionary"
|
|
|
18 |
|
19 |
-
You will need to rename or copy appropriate files for this as mentioned below
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
20 |
|
21 |
## graph-embeddings.py
|
22 |
|
|
|
9 |
* "graph-embeddings.py": plots graph of full values of two embeddings
|
10 |
|
11 |
|
12 |
+
## (clipmodel,cliptextmodel)-calculate-distances.py
|
13 |
|
14 |
Loads the generated embeddings, reads in a word, calculates "distance" to every
|
15 |
embedding, and then shows the closest "neighbours".
|
16 |
|
17 |
+
To run this requires the files "embeddings.safetensors" and "dictionary",
|
18 |
+
in matching format
|
19 |
|
20 |
+
You will need to rename or copy appropriate files for this as mentioned below.
|
21 |
+
|
22 |
+
Note that SD models use cliptextmodel, NOT clipmodel
|
23 |
+
|
24 |
+
## graph-textmodels.py
|
25 |
+
|
26 |
+
Shows the difference between the same word, embedded by CLIPTextModel
|
27 |
+
vs CLIPModel
|
28 |
|
29 |
## graph-embeddings.py
|
30 |
|
graph-textmodels.py
ADDED
@@ -0,0 +1,113 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#!/bin/env python
|
2 |
+
|
3 |
+
"""
|
4 |
+
Input a single word, and it will graph it,
|
5 |
+
as embedded by CLIPModel vs CLIPTextModel
|
6 |
+
|
7 |
+
It will then print out the "distance" between the two,
|
8 |
+
and then show you a coordinate graph
|
9 |
+
|
10 |
+
You will want to zoom in to actually see the differences, usually
|
11 |
+
|
12 |
+
"""
|
13 |
+
|
14 |
+
|
15 |
+
import sys
|
16 |
+
import json
|
17 |
+
import torch
|
18 |
+
from transformers import CLIPProcessor,CLIPModel,CLIPTextModel
|
19 |
+
import logging
|
20 |
+
# Turn off stupid mesages from CLIPModel.load
|
21 |
+
logging.disable(logging.WARNING)
|
22 |
+
|
23 |
+
import PyQt5
|
24 |
+
import matplotlib
|
25 |
+
matplotlib.use('QT5Agg') # Set the backend to TkAgg
|
26 |
+
|
27 |
+
import matplotlib.pyplot as plt
|
28 |
+
|
29 |
+
|
30 |
+
clipsrc="openai/clip-vit-large-patch14"
|
31 |
+
|
32 |
+
overlaymodel="text_encoder.bin"
|
33 |
+
overlaymodel2="text_encoder2.bin"
|
34 |
+
|
35 |
+
processor=None
|
36 |
+
clipmodel=None
|
37 |
+
cliptextmodel=None
|
38 |
+
|
39 |
+
device=torch.device("cuda")
|
40 |
+
|
41 |
+
print("loading processor from "+clipsrc,file=sys.stderr)
|
42 |
+
processor = CLIPProcessor.from_pretrained(clipsrc)
|
43 |
+
print("done",file=sys.stderr)
|
44 |
+
|
45 |
+
def clipmodel_one_time(text):
|
46 |
+
global clipmodel
|
47 |
+
if clipmodel == None:
|
48 |
+
print("loading CLIPModel from "+clipsrc,file=sys.stderr)
|
49 |
+
clipmodel = CLIPModel.from_pretrained(clipsrc)
|
50 |
+
clipmodel = clipmodel.to(device)
|
51 |
+
print("done",file=sys.stderr)
|
52 |
+
|
53 |
+
inputs = processor(text=text, return_tensors="pt")
|
54 |
+
inputs.to(device)
|
55 |
+
with torch.no_grad():
|
56 |
+
text_features = clipmodel.get_text_features(**inputs)
|
57 |
+
return text_features
|
58 |
+
#shape = (1,768)
|
59 |
+
|
60 |
+
|
61 |
+
def cliptextmodel_one_time(text):
|
62 |
+
global cliptextmodel
|
63 |
+
if cliptextmodel == None:
|
64 |
+
print("loading CLIPTextModel from "+clipsrc,file=sys.stderr)
|
65 |
+
cliptextmodel = CLIPTextModel.from_pretrained(clipsrc)
|
66 |
+
cliptextmodel = cliptextmodel.to(device)
|
67 |
+
print("done",file=sys.stderr)
|
68 |
+
inputs = processor(text=text, return_tensors="pt")
|
69 |
+
inputs.to(device)
|
70 |
+
with torch.no_grad():
|
71 |
+
outputs = cliptextmodel(**inputs)
|
72 |
+
embeddings = outputs.pooler_output
|
73 |
+
return embeddings
|
74 |
+
# shape is (1,768)
|
75 |
+
|
76 |
+
def print_distance(emb1,emb2):
|
77 |
+
targetdistance = torch.norm( emb1 - emb2)
|
78 |
+
print("DISTANCE:",targetdistance)
|
79 |
+
|
80 |
+
|
81 |
+
|
82 |
+
def prompt_for_word():
|
83 |
+
fig, ax = plt.subplots()
|
84 |
+
|
85 |
+
text1 = input("Word or prompt: ")
|
86 |
+
if text1 == "q":
|
87 |
+
exit(0)
|
88 |
+
|
89 |
+
print("generating embeddings for each now")
|
90 |
+
|
91 |
+
emb1 = clipmodel_one_time(text1)[0]
|
92 |
+
graph1=emb1.tolist()
|
93 |
+
ax.plot(graph1, label="clipmodel")
|
94 |
+
|
95 |
+
emb2 = cliptextmodel_one_time(text1)[0]
|
96 |
+
graph2=emb2.tolist()
|
97 |
+
ax.plot(graph2, label="cliptextmodel")
|
98 |
+
|
99 |
+
print_distance(emb1,emb2)
|
100 |
+
|
101 |
+
# Add labels, title, and legend
|
102 |
+
#ax.set_xlabel('Index')
|
103 |
+
ax.set_ylabel('Values')
|
104 |
+
ax.set_title('Graph embedding from std libs')
|
105 |
+
ax.legend()
|
106 |
+
|
107 |
+
# Display the graph
|
108 |
+
print("Pulling up the graph. To calculate more distances, close graph")
|
109 |
+
plt.show()
|
110 |
+
# Dont know why plt.show only works once !
|
111 |
+
|
112 |
+
while True:
|
113 |
+
prompt_for_word()
|