Upload calculate-2distance.py
Browse files- calculate-2distance.py +101 -0
calculate-2distance.py
ADDED
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#!/bin/env python
|
2 |
+
|
3 |
+
"""
|
4 |
+
Plan:
|
5 |
+
Read in "dictionary" for list of words
|
6 |
+
Read in pre-calculated "proper" embedding for each word from
|
7 |
+
safetensor file named "embeddings.safetensors"
|
8 |
+
Prompt user for two words from the list
|
9 |
+
(but may also be off the list, or a phrase)
|
10 |
+
Print out Euclidean distance between the two
|
11 |
+
|
12 |
+
(the point of the dictionary is that it can make loading super fast for known words)
|
13 |
+
|
14 |
+
"""
|
15 |
+
|
16 |
+
|
17 |
+
import sys
|
18 |
+
import json
|
19 |
+
import torch
|
20 |
+
from safetensors import safe_open
|
21 |
+
|
22 |
+
import numpy
|
23 |
+
|
24 |
+
from transformers import CLIPProcessor,CLIPModel
|
25 |
+
|
26 |
+
clipsrc="openai/clip-vit-large-patch14"
|
27 |
+
processor=None
|
28 |
+
model=None
|
29 |
+
|
30 |
+
device=torch.device("cuda")
|
31 |
+
|
32 |
+
|
33 |
+
def init():
|
34 |
+
global processor
|
35 |
+
global model
|
36 |
+
# Load the processor and model
|
37 |
+
print("loading processor from "+clipsrc,file=sys.stderr)
|
38 |
+
processor = CLIPProcessor.from_pretrained(clipsrc)
|
39 |
+
print("done",file=sys.stderr)
|
40 |
+
print("loading model from "+clipsrc,file=sys.stderr)
|
41 |
+
model = CLIPModel.from_pretrained(clipsrc)
|
42 |
+
print("done",file=sys.stderr)
|
43 |
+
|
44 |
+
model = model.to(device)
|
45 |
+
|
46 |
+
|
47 |
+
|
48 |
+
embed_file="embeddings.safetensors"
|
49 |
+
|
50 |
+
device=torch.device("cuda")
|
51 |
+
|
52 |
+
print("read in words from dictionary now",file=sys.stderr)
|
53 |
+
with open("dictionary","r") as f:
|
54 |
+
tokendict = f.readlines()
|
55 |
+
wordlist = [token.strip() for token in tokendict] # Remove trailing newlines
|
56 |
+
print(len(wordlist),"lines read")
|
57 |
+
|
58 |
+
print("read in embeddings now",file=sys.stderr)
|
59 |
+
model = safe_open(embed_file,framework="pt",device="cuda")
|
60 |
+
embs=model.get_tensor("embeddings")
|
61 |
+
embs.to(device)
|
62 |
+
print("Shape of loaded embeds =",embs.shape)
|
63 |
+
|
64 |
+
def standard_embed_calc(text):
|
65 |
+
if processor == None:
|
66 |
+
init()
|
67 |
+
|
68 |
+
inputs = processor(text=text, return_tensors="pt")
|
69 |
+
inputs.to(device)
|
70 |
+
with torch.no_grad():
|
71 |
+
text_features = model.get_text_features(**inputs)
|
72 |
+
embedding = text_features[0]
|
73 |
+
return embedding
|
74 |
+
|
75 |
+
|
76 |
+
def print_distance(emb1,emb2):
|
77 |
+
targetdistance = torch.norm( emb1 - emb2)
|
78 |
+
print("DISTANCE:",targetdistance)
|
79 |
+
|
80 |
+
|
81 |
+
# return embed of target word.
|
82 |
+
# pull from dictionary, or do full calc
|
83 |
+
def find_word(targetword):
|
84 |
+
try:
|
85 |
+
targetindex=wordlist.index(targetword)
|
86 |
+
targetemb=embs[targetindex]
|
87 |
+
return targetemb
|
88 |
+
return
|
89 |
+
except ValueError:
|
90 |
+
print(targetword,"not found in cache")
|
91 |
+
print("Now doing lookup with full calc embed")
|
92 |
+
targetemb=standard_embed_calc(targetword)
|
93 |
+
return targetemb
|
94 |
+
|
95 |
+
|
96 |
+
while True:
|
97 |
+
input_text1=input("Input a word1(or phrase) now:")
|
98 |
+
input_text2=input("Input word2 now:")
|
99 |
+
emb1=find_word(input_text1)
|
100 |
+
emb2=find_word(input_text2)
|
101 |
+
print_distance(emb1,emb2)
|