Spaces:
Sleeping
Sleeping
feat: support pdf docs
Browse files- app.py +120 -49
- requirements.txt +1 -0
app.py
CHANGED
@@ -1,64 +1,135 @@
|
|
1 |
import streamlit as st
|
2 |
-
|
|
|
3 |
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
|
4 |
|
5 |
checkpoint = "facebook/bart-large-cnn"
|
6 |
|
|
|
7 |
@st.cache_resource
|
8 |
def load_model():
|
9 |
model = AutoModelForSeq2SeqLM.from_pretrained(checkpoint)
|
10 |
return model
|
11 |
|
|
|
12 |
@st.cache_resource
|
13 |
def load_tokenizer():
|
14 |
tokenizer = AutoTokenizer.from_pretrained(checkpoint)
|
15 |
return tokenizer
|
16 |
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import streamlit as st
|
2 |
+
import os
|
3 |
+
import PyPDF2
|
4 |
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
|
5 |
|
6 |
checkpoint = "facebook/bart-large-cnn"
|
7 |
|
8 |
+
|
9 |
@st.cache_resource
|
10 |
def load_model():
|
11 |
model = AutoModelForSeq2SeqLM.from_pretrained(checkpoint)
|
12 |
return model
|
13 |
|
14 |
+
|
15 |
@st.cache_resource
|
16 |
def load_tokenizer():
|
17 |
tokenizer = AutoTokenizer.from_pretrained(checkpoint)
|
18 |
return tokenizer
|
19 |
|
20 |
+
|
21 |
+
def load_text_file(file):
|
22 |
+
bytes_data = file.getvalue()
|
23 |
+
text = bytes_data.decode("utf-8")
|
24 |
+
return text
|
25 |
+
|
26 |
+
|
27 |
+
def load_pdf_file(file):
|
28 |
+
pdf_reader = PyPDF2.PdfReader(file)
|
29 |
+
pdf_text = ""
|
30 |
+
for page_num in range(len(pdf_reader.pages)):
|
31 |
+
pdf_text += pdf_reader.pages[page_num].extract_text() or ""
|
32 |
+
return pdf_text
|
33 |
+
|
34 |
+
|
35 |
+
def split_text_into_chunks(text, max_chunk_length):
|
36 |
+
chunks = []
|
37 |
+
current_chunk = ""
|
38 |
+
|
39 |
+
for word in text.split():
|
40 |
+
if len(current_chunk) + len(word) + 1 <= max_chunk_length:
|
41 |
+
current_chunk += word + " "
|
42 |
+
else:
|
43 |
+
chunks.append(current_chunk.strip())
|
44 |
+
current_chunk = word + " "
|
45 |
+
|
46 |
+
if current_chunk:
|
47 |
+
chunks.append(current_chunk.strip())
|
48 |
+
|
49 |
+
return chunks
|
50 |
+
|
51 |
+
|
52 |
+
def main():
|
53 |
+
model = load_model()
|
54 |
+
print("Model's maximum sequence length:", model.config.max_position_embeddings)
|
55 |
+
|
56 |
+
tokenizer = load_tokenizer()
|
57 |
+
print("Tokenizer's maximum sequence length:", tokenizer.model_max_length)
|
58 |
+
|
59 |
+
st.title("Summarisation Tool")
|
60 |
+
st.write(
|
61 |
+
f"Performs basic summarisation of text and audio using the '{checkpoint}' model."
|
62 |
+
)
|
63 |
+
|
64 |
+
st.sidebar.title("Options")
|
65 |
+
summary_balance = st.sidebar.select_slider(
|
66 |
+
"Output Summarisation Detail:",
|
67 |
+
options=["concise", "balanced", "full"],
|
68 |
+
value="balanced",
|
69 |
+
)
|
70 |
+
|
71 |
+
textTab, docTab, audioTab = st.tabs(["Plain Text", "Text Document", "Audio File"])
|
72 |
+
|
73 |
+
with textTab:
|
74 |
+
sentence = st.text_area(
|
75 |
+
"Paste text to be summarised:",
|
76 |
+
help="Paste text into text area and hit Summarise button",
|
77 |
+
height=300,
|
78 |
+
)
|
79 |
+
st.write(f"{len(sentence)} characters and {len(sentence.split())} words")
|
80 |
+
|
81 |
+
with docTab:
|
82 |
+
uploaded_file = st.file_uploader("Select a file to be summarised:")
|
83 |
+
if uploaded_file is not None:
|
84 |
+
file_name = os.path.basename(uploaded_file.name)
|
85 |
+
_, file_ext = os.path.splitext(file_name)
|
86 |
+
if "pdf" in file_ext:
|
87 |
+
sentence = load_pdf_file(uploaded_file)
|
88 |
+
else:
|
89 |
+
sentence = load_text_file(uploaded_file)
|
90 |
+
st.write(f"{len(sentence)} characters and {len(sentence.split())} words")
|
91 |
+
st.write(sentence)
|
92 |
+
|
93 |
+
with audioTab:
|
94 |
+
st.text("Yet to be implemented...")
|
95 |
+
|
96 |
+
button = st.button("Summarise")
|
97 |
+
st.divider()
|
98 |
+
|
99 |
+
with st.spinner("Generating Summary..."):
|
100 |
+
if button and sentence:
|
101 |
+
chunks = split_text_into_chunks(sentence, 500)
|
102 |
+
print(chunks)
|
103 |
+
|
104 |
+
text_words = len(sentence.split())
|
105 |
+
if summary_balance == "concise":
|
106 |
+
min_multiplier = text_words * 0.1
|
107 |
+
max_multiplier = text_words * 0.3
|
108 |
+
elif summary_balance == "full":
|
109 |
+
min_multiplier = text_words * 0.5
|
110 |
+
max_multiplier = text_words * 0.8
|
111 |
+
elif summary_balance == "balanced":
|
112 |
+
min_multiplier = text_words * 0.2
|
113 |
+
max_multiplier = text_words * 0.5
|
114 |
+
|
115 |
+
print(f"min tokens {int(min_multiplier)}, max tokens {int(max_multiplier)}")
|
116 |
+
inputs = tokenizer(
|
117 |
+
chunks,
|
118 |
+
max_length=model.config.max_position_embeddings,
|
119 |
+
return_tensors="pt",
|
120 |
+
truncation=True,
|
121 |
+
padding=True,
|
122 |
+
)
|
123 |
+
summary_ids = model.generate(
|
124 |
+
inputs["input_ids"],
|
125 |
+
min_new_tokens=int(min_multiplier),
|
126 |
+
max_new_tokens=int(max_multiplier),
|
127 |
+
do_sample=False,
|
128 |
+
)
|
129 |
+
summary = tokenizer.decode(summary_ids[0], skip_special_tokens=True)
|
130 |
+
st.write(summary)
|
131 |
+
st.write(f"{len(summary)} characters and {len(summary.split())} words")
|
132 |
+
|
133 |
+
|
134 |
+
if __name__ == "__main__":
|
135 |
+
main()
|
requirements.txt
CHANGED
@@ -5,3 +5,4 @@ torch
|
|
5 |
torchvision
|
6 |
torchaudio
|
7 |
transformers
|
|
|
|
5 |
torchvision
|
6 |
torchaudio
|
7 |
transformers
|
8 |
+
PyPDF2
|