Spaces:
Build error
Build error
Commit
·
5bdaeed
1
Parent(s):
3e95584
Create api.py
Browse files
api.py
ADDED
@@ -0,0 +1,177 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import re
|
3 |
+
import shutil
|
4 |
+
import urllib.request
|
5 |
+
from pathlib import Path
|
6 |
+
from tempfile import NamedTemporaryFile
|
7 |
+
|
8 |
+
import fitz
|
9 |
+
import numpy as np
|
10 |
+
import openai
|
11 |
+
import tensorflow_hub as hub
|
12 |
+
from fastapi import UploadFile
|
13 |
+
from lcserve import serving
|
14 |
+
from sklearn.neighbors import NearestNeighbors
|
15 |
+
|
16 |
+
|
17 |
+
recommender = None
|
18 |
+
|
19 |
+
|
20 |
+
def download_pdf(url, output_path):
|
21 |
+
urllib.request.urlretrieve(url, output_path)
|
22 |
+
|
23 |
+
|
24 |
+
def preprocess(text):
|
25 |
+
text = text.replace('\n', ' ')
|
26 |
+
text = re.sub('\s+', ' ', text)
|
27 |
+
return text
|
28 |
+
|
29 |
+
|
30 |
+
def pdf_to_text(path, start_page=1, end_page=None):
|
31 |
+
doc = fitz.open(path)
|
32 |
+
total_pages = doc.page_count
|
33 |
+
|
34 |
+
if end_page is None:
|
35 |
+
end_page = total_pages
|
36 |
+
|
37 |
+
text_list = []
|
38 |
+
|
39 |
+
for i in range(start_page - 1, end_page):
|
40 |
+
text = doc.load_page(i).get_text("text")
|
41 |
+
text = preprocess(text)
|
42 |
+
text_list.append(text)
|
43 |
+
|
44 |
+
doc.close()
|
45 |
+
return text_list
|
46 |
+
|
47 |
+
|
48 |
+
def text_to_chunks(texts, word_length=150, start_page=1):
|
49 |
+
text_toks = [t.split(' ') for t in texts]
|
50 |
+
chunks = []
|
51 |
+
|
52 |
+
for idx, words in enumerate(text_toks):
|
53 |
+
for i in range(0, len(words), word_length):
|
54 |
+
chunk = words[i : i + word_length]
|
55 |
+
if (
|
56 |
+
(i + word_length) > len(words)
|
57 |
+
and (len(chunk) < word_length)
|
58 |
+
and (len(text_toks) != (idx + 1))
|
59 |
+
):
|
60 |
+
text_toks[idx + 1] = chunk + text_toks[idx + 1]
|
61 |
+
continue
|
62 |
+
chunk = ' '.join(chunk).strip()
|
63 |
+
chunk = f'[Page no. {idx+start_page}]' + ' ' + '"' + chunk + '"'
|
64 |
+
chunks.append(chunk)
|
65 |
+
return chunks
|
66 |
+
|
67 |
+
|
68 |
+
class SemanticSearch:
|
69 |
+
def __init__(self):
|
70 |
+
self.use = hub.load('https://tfhub.dev/google/universal-sentence-encoder/4')
|
71 |
+
self.fitted = False
|
72 |
+
|
73 |
+
def fit(self, data, batch=1000, n_neighbors=5):
|
74 |
+
self.data = data
|
75 |
+
self.embeddings = self.get_text_embedding(data, batch=batch)
|
76 |
+
n_neighbors = min(n_neighbors, len(self.embeddings))
|
77 |
+
self.nn = NearestNeighbors(n_neighbors=n_neighbors)
|
78 |
+
self.nn.fit(self.embeddings)
|
79 |
+
self.fitted = True
|
80 |
+
|
81 |
+
def __call__(self, text, return_data=True):
|
82 |
+
inp_emb = self.use([text])
|
83 |
+
neighbors = self.nn.kneighbors(inp_emb, return_distance=False)[0]
|
84 |
+
|
85 |
+
if return_data:
|
86 |
+
return [self.data[i] for i in neighbors]
|
87 |
+
else:
|
88 |
+
return neighbors
|
89 |
+
|
90 |
+
def get_text_embedding(self, texts, batch=1000):
|
91 |
+
embeddings = []
|
92 |
+
for i in range(0, len(texts), batch):
|
93 |
+
text_batch = texts[i : (i + batch)]
|
94 |
+
emb_batch = self.use(text_batch)
|
95 |
+
embeddings.append(emb_batch)
|
96 |
+
embeddings = np.vstack(embeddings)
|
97 |
+
return embeddings
|
98 |
+
|
99 |
+
|
100 |
+
def load_recommender(path, start_page=1):
|
101 |
+
global recommender
|
102 |
+
if recommender is None:
|
103 |
+
recommender = SemanticSearch()
|
104 |
+
|
105 |
+
texts = pdf_to_text(path, start_page=start_page)
|
106 |
+
chunks = text_to_chunks(texts, start_page=start_page)
|
107 |
+
recommender.fit(chunks)
|
108 |
+
return 'Corpus Loaded.'
|
109 |
+
|
110 |
+
|
111 |
+
def generate_text(openAI_key, prompt, engine="text-davinci-003"):
|
112 |
+
openai.api_key = openAI_key
|
113 |
+
try:
|
114 |
+
completions = openai.Completion.create(
|
115 |
+
engine=engine,
|
116 |
+
prompt=prompt,
|
117 |
+
max_tokens=512,
|
118 |
+
n=1,
|
119 |
+
stop=None,
|
120 |
+
temperature=0.7,
|
121 |
+
)
|
122 |
+
message = completions.choices[0].text
|
123 |
+
except Exception as e:
|
124 |
+
message = f'API Error: {str(e)}'
|
125 |
+
return message
|
126 |
+
|
127 |
+
|
128 |
+
def generate_answer(question, openAI_key):
|
129 |
+
topn_chunks = recommender(question)
|
130 |
+
prompt = ""
|
131 |
+
prompt += 'search results:\n\n'
|
132 |
+
for c in topn_chunks:
|
133 |
+
prompt += c + '\n\n'
|
134 |
+
|
135 |
+
prompt += (
|
136 |
+
"Instructions: Compose a comprehensive reply to the query using the search results given. "
|
137 |
+
"Cite each reference using [ Page Number] notation (every result has this number at the beginning). "
|
138 |
+
"Citation should be done at the end of each sentence. If the search results mention multiple subjects "
|
139 |
+
"with the same name, create separate answers for each. Only include information found in the results and "
|
140 |
+
"don't add any additional information. Make sure the answer is correct and don't output false content. "
|
141 |
+
"If the text does not relate to the query, simply state 'Text Not Found in PDF'. Ignore outlier "
|
142 |
+
"search results which has nothing to do with the question. Only answer what is asked. The "
|
143 |
+
"answer should be short and concise. Answer step-by-step. \n\nQuery: {question}\nAnswer: "
|
144 |
+
)
|
145 |
+
|
146 |
+
prompt += f"Query: {question}\nAnswer:"
|
147 |
+
answer = generate_text(openAI_key, prompt, "text-davinci-003")
|
148 |
+
return answer
|
149 |
+
|
150 |
+
|
151 |
+
def load_openai_key() -> str:
|
152 |
+
key = os.environ.get("OPENAI_API_KEY")
|
153 |
+
if key is None:
|
154 |
+
raise ValueError(
|
155 |
+
"[ERROR]: Please pass your OPENAI_API_KEY. Get your key here : https://platform.openai.com/account/api-keys"
|
156 |
+
)
|
157 |
+
return key
|
158 |
+
|
159 |
+
|
160 |
+
@serving
|
161 |
+
def ask_url(url: str, question: str):
|
162 |
+
download_pdf(url, 'corpus.pdf')
|
163 |
+
load_recommender('corpus.pdf')
|
164 |
+
openAI_key = load_openai_key()
|
165 |
+
return generate_answer(question, openAI_key)
|
166 |
+
|
167 |
+
|
168 |
+
@serving
|
169 |
+
async def ask_file(file: UploadFile, question: str) -> str:
|
170 |
+
suffix = Path(file.filename).suffix
|
171 |
+
with NamedTemporaryFile(delete=False, suffix=suffix) as tmp:
|
172 |
+
shutil.copyfileobj(file.file, tmp)
|
173 |
+
tmp_path = Path(tmp.name)
|
174 |
+
|
175 |
+
load_recommender(str(tmp_path))
|
176 |
+
openAI_key = load_openai_key()
|
177 |
+
return generate_answer(question, openAI_key)
|