Upload generate-embeddingXL.py
Browse files- generate-embeddingXL.py +99 -0
generate-embeddingXL.py
ADDED
@@ -0,0 +1,99 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#!/bin/env python
|
2 |
+
|
3 |
+
""" Work in progress
|
4 |
+
|
5 |
+
Similar to generate-embedding.py, but outputs in the format
|
6 |
+
that SDXL models expect. I hope.
|
7 |
+
|
8 |
+
Also tries to load the SDXL base text encoder specifically.
|
9 |
+
Requires you populate the two paths mentioned immediately below this comment section.
|
10 |
+
|
11 |
+
You can get them from:
|
12 |
+
https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/tree/main/text_encoder_2
|
13 |
+
|
14 |
+
(rename diffusion_pytorch_model.safetensors to text_encoder_xl.safetensors)
|
15 |
+
|
16 |
+
|
17 |
+
Plan:
|
18 |
+
Take input for a single word or phrase.
|
19 |
+
Save out calculations, to "generatedXL.safetensors"
|
20 |
+
|
21 |
+
Note that you can generate an embedding from two words, or even more
|
22 |
+
|
23 |
+
I could also include a "clip_l" key, but..
|
24 |
+
Meh.
|
25 |
+
"""
|
26 |
+
|
27 |
+
model_path = "text_encoder_xl.safetensors"
|
28 |
+
model_config = "text_encoder_2_config.json"
|
29 |
+
|
30 |
+
import sys
|
31 |
+
import torch
|
32 |
+
from transformers import CLIPProcessor, CLIPTextModel, CLIPTextModelWithProjection
|
33 |
+
from safetensors.torch import save_file
|
34 |
+
|
35 |
+
# 1. Load the pretrained model
|
36 |
+
# Note that it doesnt like a leading "/" in the name!!
|
37 |
+
|
38 |
+
|
39 |
+
model=None
|
40 |
+
processor=None
|
41 |
+
|
42 |
+
device=torch.device("cuda")
|
43 |
+
|
44 |
+
# Note the default, required 2 pathnames
|
45 |
+
def initXLCLIPmodel():
|
46 |
+
global model
|
47 |
+
print("loading",model_path)
|
48 |
+
model = CLIPTextModelWithProjection.from_pretrained(model_path,config=model_config,local_files_only=True,use_safetensors=True)
|
49 |
+
model.to(device)
|
50 |
+
|
51 |
+
# a bit wierd, but SDXL seems to still use this tokeninzer
|
52 |
+
def initCLIPprocessor():
|
53 |
+
global processor
|
54 |
+
CLIPname= "openai/clip-vit-large-patch14"
|
55 |
+
print("getting processor from",CLIPname)
|
56 |
+
processor = CLIPProcessor.from_pretrained(CLIPname)
|
57 |
+
|
58 |
+
def embed_from_text(text):
|
59 |
+
global processor,model
|
60 |
+
if processor == None:
|
61 |
+
initCLIPprocessor()
|
62 |
+
initXLCLIPmodel()
|
63 |
+
print("getting tokens")
|
64 |
+
inputs = processor(text=text, return_tensors="pt")
|
65 |
+
inputs.to(device)
|
66 |
+
|
67 |
+
print("getting embeddings?")
|
68 |
+
outputs = model(**inputs)
|
69 |
+
print("finalizing")
|
70 |
+
embeddings = outputs.text_embeds
|
71 |
+
return embeddings
|
72 |
+
|
73 |
+
|
74 |
+
|
75 |
+
##########################################
|
76 |
+
|
77 |
+
word = input("type a phrase to generate an embedding for: ")
|
78 |
+
|
79 |
+
emb = embed_from_text(word)
|
80 |
+
#embs=emb.unsqueeze(0) # stupid matrix magic to make it the required shape
|
81 |
+
embs=emb
|
82 |
+
|
83 |
+
print("Shape of result = ",embs.shape)
|
84 |
+
# Note that programs like shapes such as
|
85 |
+
# torch.Size([1, 768])
|
86 |
+
|
87 |
+
output = "generatedXL.safetensors"
|
88 |
+
# if single word used, then rename output file
|
89 |
+
if all(char.isalpha() for char in word):
|
90 |
+
output=f"{word}XL.safetensors"
|
91 |
+
print(f"Saving to {output}...")
|
92 |
+
save_file({"clip_g": embs}, output)
|
93 |
+
|
94 |
+
# technically we are saving a shape ([1][1280])
|
95 |
+
# whereas official XL embeddings files, are
|
96 |
+
# (clip_g) shape ([8][1280])
|
97 |
+
# (clip_l) shape ([8][768])
|
98 |
+
|
99 |
+
|