Sortoite commited on
Commit
647f505
1 Parent(s): 01611c5

Create app.py

Browse files
Files changed (1) hide show
  1. app.py +238 -0
app.py ADDED
@@ -0,0 +1,238 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import urllib.request
2
+ import fitz
3
+ import re
4
+ import numpy as np
5
+ import tensorflow_hub as hub
6
+ import openai
7
+ import gradio as gr
8
+ import os
9
+ from sklearn.neighbors import NearestNeighbors
10
+
11
+ def download_pdf(url, output_path):
12
+ urllib.request.urlretrieve(url, output_path)
13
+
14
+
15
+ def preprocess(text):
16
+ text = text.replace('\n', ' ')
17
+ text = re.sub('\s+', ' ', text)
18
+ return text
19
+
20
+
21
+ def pdf_to_text(path, start_page=1, end_page=None):
22
+ doc = fitz.open(path)
23
+ total_pages = doc.page_count
24
+
25
+ if end_page is None:
26
+ end_page = total_pages
27
+
28
+ text_list = []
29
+
30
+ for i in range(start_page-1, end_page):
31
+ text = doc.load_page(i).get_text("text")
32
+ text = preprocess(text)
33
+ text_list.append(text)
34
+
35
+ doc.close()
36
+ return text_list
37
+
38
+
39
+ def text_to_chunks(texts, word_length=150, start_page=1):
40
+ text_toks = [t.split(' ') for t in texts]
41
+ page_nums = []
42
+ chunks = []
43
+
44
+ for idx, words in enumerate(text_toks):
45
+ for i in range(0, len(words), word_length):
46
+ chunk = words[i:i+word_length]
47
+ if (i+word_length) > len(words) and (len(chunk) < word_length) and (
48
+ len(text_toks) != (idx+1)):
49
+ text_toks[idx+1] = chunk + text_toks[idx+1]
50
+ continue
51
+ chunk = ' '.join(chunk).strip()
52
+ chunk = f'[{idx+start_page}]' + ' ' + '"' + chunk + '"'
53
+ chunks.append(chunk)
54
+ return chunks
55
+
56
+
57
+ class SemanticSearch:
58
+
59
+ def __init__(self):
60
+ self.use = hub.load('https://tfhub.dev/google/universal-sentence-encoder/4')
61
+ self.fitted = False
62
+
63
+
64
+ def fit(self, data, batch=1000, n_neighbors=5):
65
+ self.data = data
66
+ self.embeddings = self.get_text_embedding(data, batch=batch)
67
+ n_neighbors = min(n_neighbors, len(self.embeddings))
68
+ self.nn = NearestNeighbors(n_neighbors=n_neighbors)
69
+ self.nn.fit(self.embeddings)
70
+ self.fitted = True
71
+
72
+
73
+ def __call__(self, text, return_data=True):
74
+ inp_emb = self.use([text])
75
+ neighbors = self.nn.kneighbors(inp_emb, return_distance=False)[0]
76
+
77
+ if return_data:
78
+ return [self.data[i] for i in neighbors]
79
+ else:
80
+ return neighbors
81
+
82
+
83
+ def get_text_embedding(self, texts, batch=1000):
84
+ embeddings = []
85
+ for i in range(0, len(texts), batch):
86
+ text_batch = texts[i:(i+batch)]
87
+ emb_batch = self.use(text_batch)
88
+ embeddings.append(emb_batch)
89
+ embeddings = np.vstack(embeddings)
90
+ return embeddings
91
+
92
+
93
+
94
+ #def load_recommender(path, start_page=1):
95
+ # global recommender
96
+ # texts = pdf_to_text(path, start_page=start_page)
97
+ # chunks = text_to_chunks(texts, start_page=start_page)
98
+ # recommender.fit(chunks)
99
+ # return 'Corpus Loaded.'
100
+
101
+ # The modified function generates embeddings based on PDF file name and page number and checks if the embeddings file exists before loading or generating it.
102
+ def load_recommender(path, start_page=1):
103
+ global recommender
104
+ pdf_file = os.path.basename(path)
105
+ embeddings_file = f"{pdf_file}_{start_page}.npy"
106
+
107
+ if os.path.isfile(embeddings_file):
108
+ embeddings = np.load(embeddings_file)
109
+ recommender.embeddings = embeddings
110
+ recommender.fitted = True
111
+ return "Embeddings loaded from file"
112
+
113
+ texts = pdf_to_text(path, start_page=start_page)
114
+ chunks = text_to_chunks(texts, start_page=start_page)
115
+ recommender.fit(chunks)
116
+ np.save(embeddings_file, recommender.embeddings)
117
+ return 'Corpus Loaded.'
118
+
119
+
120
+
121
+ def generate_text(openAI_key,prompt, engine="text-davinci-003"):
122
+ openai.api_key = openAI_key
123
+ completions = openai.Completion.create(
124
+ engine=engine,
125
+ prompt=prompt,
126
+ max_tokens=512,
127
+ n=1,
128
+ stop=None,
129
+ temperature=0.7,
130
+ )
131
+ message = completions.choices[0].text
132
+ return message
133
+
134
+
135
+ def generate_answer(question,openAI_key):
136
+ topn_chunks = recommender(question)
137
+ prompt = ""
138
+ prompt += 'search results:\n\n'
139
+ for c in topn_chunks:
140
+ prompt += c + '\n\n'
141
+
142
+ prompt += "Instructions: Compose a comprehensive reply to the query using the search results given. "\
143
+ "Cite each reference using [number] notation (every result has this number at the beginning). "\
144
+ "Citation should be done at the end of each sentence. If the search results mention multiple subjects "\
145
+ "with the same name, create separate answers for each. Only include information found in the results and "\
146
+ "don't add any additional information. Make sure the answer is correct and don't output false content. "\
147
+ "If the text does not relate to the query, simply state 'Found Nothing'. Ignore outlier "\
148
+ "search results which has nothing to do with the question. Only answer what is asked. The "\
149
+ "answer should be short and concise.\n\nQuery: {question}\nAnswer: "
150
+
151
+ prompt += f"Query: {question}\nAnswer:"
152
+ answer = generate_text(openAI_key, prompt,"text-davinci-003")
153
+ return answer
154
+
155
+
156
+ def question_answer(url, file, question,openAI_key):
157
+ if openAI_key.strip()=='':
158
+ return '[ERROR]: Please enter you Open AI Key. Get your key here : https://platform.openai.com/account/api-keys'
159
+ if url.strip() == '' and file == None:
160
+ return '[ERROR]: Both URL and PDF is empty. Provide atleast one.'
161
+
162
+ if url.strip() != '' and file != None:
163
+ return '[ERROR]: Both URL and PDF is provided. Please provide only one (eiter URL or PDF).'
164
+
165
+ if url.strip() != '':
166
+ glob_url = url
167
+ download_pdf(glob_url, 'corpus.pdf')
168
+ load_recommender('corpus.pdf')
169
+
170
+ else:
171
+ old_file_name = file.name
172
+ file_name = file.name
173
+ file_name = file_name[:-12] + file_name[-4:]
174
+ os.rename(old_file_name, file_name)
175
+ load_recommender(file_name)
176
+
177
+ if question.strip() == '':
178
+ return '[ERROR]: Question field is empty'
179
+
180
+ return generate_answer(question,openAI_key)
181
+
182
+
183
+ recommender = SemanticSearch()
184
+
185
+ title = 'PDF GPT'
186
+ description = """ What is PDF GPT ?
187
+ 1. The problem is that Open AI has a 4K token limit and cannot take an entire PDF file as input. Additionally, it sometimes returns irrelevant responses due to poor embeddings. ChatGPT cannot directly talk to external data. The solution is PDF GPT, which allows you to chat with an uploaded PDF file using GPT functionalities. The application breaks the document into smaller chunks and generates embeddings using a powerful Deep Averaging Network Encoder. A semantic search is performed on your query, and the top relevant chunks are used to generate a response.
188
+ 2. The returned response can even cite the page number in square brackets([]) where the information is located, adding credibility to the responses and helping to locate pertinent information quickly. The Responses are much better than the naive responses by Open AI."""
189
+
190
+ with gr.Blocks() as demo:
191
+
192
+ gr.Markdown(f'<center><h1>{title}</h1></center>')
193
+ gr.Markdown(description)
194
+
195
+ with gr.Row():
196
+
197
+ with gr.Group():
198
+ gr.Markdown(f'<p style="text-align:center">Get your Open AI API key <a href="https://platform.openai.com/account/api-keys">here</a></p>')
199
+ openAI_key=gr.Textbox(label='Enter your OpenAI API key here')
200
+ url = gr.Textbox(label='Enter PDF URL here')
201
+ gr.Markdown("<center><h4>OR<h4></center>")
202
+ file = gr.File(label='Upload your PDF/ Research Paper / Book here', file_types=['.pdf'])
203
+ question = gr.Textbox(label='Enter your question here')
204
+ btn = gr.Button(value='Submit')
205
+ btn.style(full_width=True)
206
+
207
+ with gr.Group():
208
+ answer = gr.Textbox(label='The answer to your question is :')
209
+
210
+ btn.click(question_answer, inputs=[url, file, question,openAI_key], outputs=[answer])
211
+ #openai.api_key = os.getenv('Your_Key_Here')
212
+ demo.launch()
213
+
214
+
215
+ # import streamlit as st
216
+
217
+ # #Define the app layout
218
+ # st.markdown(f'<center><h1>{title}</h1></center>', unsafe_allow_html=True)
219
+ # st.markdown(description)
220
+
221
+ # col1, col2 = st.columns(2)
222
+
223
+ # # Define the inputs in the first column
224
+ # with col1:
225
+ # url = st.text_input('URL')
226
+ # st.markdown("<center><h6>or<h6></center>", unsafe_allow_html=True)
227
+ # file = st.file_uploader('PDF', type='pdf')
228
+ # question = st.text_input('question')
229
+ # btn = st.button('Submit')
230
+
231
+ # # Define the output in the second column
232
+ # with col2:
233
+ # answer = st.text_input('answer')
234
+
235
+ # # Define the button action
236
+ # if btn:
237
+ # answer_value = question_answer(url, file, question)
238
+ # answer.value = answer_value