Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -2,6 +2,7 @@ import tqdm
|
|
2 |
from PIL import Image
|
3 |
import hashlib
|
4 |
import torch
|
|
|
5 |
import fitz
|
6 |
import threading
|
7 |
import gradio as gr
|
@@ -18,6 +19,36 @@ import json
|
|
18 |
cache_dir = '/data/KB'
|
19 |
os.makedirs(cache_dir, exist_ok=True)
|
20 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
21 |
def get_image_md5(img: Image.Image):
|
22 |
img_byte_array = img.tobytes()
|
23 |
hash_md5 = hashlib.md5()
|
@@ -57,8 +88,8 @@ def add_pdf_gradio(pdf_file_binary, progress=gr.Progress()):
|
|
57 |
image_md5 = get_image_md5(image)
|
58 |
image_md5s.append(image_md5)
|
59 |
with torch.no_grad():
|
60 |
-
reps =
|
61 |
-
reps_list.append(reps
|
62 |
images.append(image)
|
63 |
|
64 |
for idx in range(len(images)):
|
@@ -95,7 +126,7 @@ def retrieve_gradio(knowledge_base: str, query: str, topk: int):
|
|
95 |
|
96 |
query_with_instruction = "Represent this query for retrieving relavant document: " + query
|
97 |
with torch.no_grad():
|
98 |
-
query_rep =
|
99 |
|
100 |
query_md5 = hashlib.md5(query.encode()).hexdigest()
|
101 |
|
|
|
2 |
from PIL import Image
|
3 |
import hashlib
|
4 |
import torch
|
5 |
+
import torch.nn.functional as F
|
6 |
import fitz
|
7 |
import threading
|
8 |
import gradio as gr
|
|
|
19 |
cache_dir = '/data/KB'
|
20 |
os.makedirs(cache_dir, exist_ok=True)
|
21 |
|
22 |
+
def weighted_mean_pooling(hidden, attention_mask):
|
23 |
+
attention_mask_ = attention_mask * attention_mask.cumsum(dim=1)
|
24 |
+
s = torch.sum(hidden * attention_mask_.unsqueeze(-1).float(), dim=1)
|
25 |
+
d = attention_mask_.sum(dim=1, keepdim=True).float()
|
26 |
+
reps = s / d
|
27 |
+
return reps
|
28 |
+
|
29 |
+
@torch.no_grad()
|
30 |
+
def encode(text_or_image_list):
|
31 |
+
global model, tokenizer
|
32 |
+
if (isinstance(text_or_image_list[0], str)):
|
33 |
+
inputs = {
|
34 |
+
"text": text_or_image_list,
|
35 |
+
'image': [None] * len(text_or_image_list),
|
36 |
+
'tokenizer': tokenizer
|
37 |
+
}
|
38 |
+
else:
|
39 |
+
inputs = {
|
40 |
+
"text": [''] * len(text_or_image_list),
|
41 |
+
'image': text_or_image_list,
|
42 |
+
'tokenizer': tokenizer
|
43 |
+
}
|
44 |
+
outputs = model(**inputs)
|
45 |
+
attention_mask = outputs.attention_mask
|
46 |
+
hidden = outputs.last_hidden_state
|
47 |
+
|
48 |
+
reps = weighted_mean_pooling(hidden, attention_mask)
|
49 |
+
embeddings = F.normalize(reps, p=2, dim=1).detach().cpu().numpy()
|
50 |
+
return embeddings
|
51 |
+
|
52 |
def get_image_md5(img: Image.Image):
|
53 |
img_byte_array = img.tobytes()
|
54 |
hash_md5 = hashlib.md5()
|
|
|
88 |
image_md5 = get_image_md5(image)
|
89 |
image_md5s.append(image_md5)
|
90 |
with torch.no_grad():
|
91 |
+
reps = encode([image])
|
92 |
+
reps_list.append(reps)
|
93 |
images.append(image)
|
94 |
|
95 |
for idx in range(len(images)):
|
|
|
126 |
|
127 |
query_with_instruction = "Represent this query for retrieving relavant document: " + query
|
128 |
with torch.no_grad():
|
129 |
+
query_rep = encode([query_with_instruction])
|
130 |
|
131 |
query_md5 = hashlib.md5(query.encode()).hexdigest()
|
132 |
|