Spaces:
Sleeping
Sleeping
File size: 6,476 Bytes
798614c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 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 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 |
import torch
from transformers import AutoTokenizer, AutoModelForTokenClassification, pipeline
import requests
import tqdm as t
import re
from sentence_transformers import SentenceTransformer
from sklearn.metrics.pairwise import cosine_similarity
import pytesseract
from PIL import Image
from collections import deque
device = "cuda" if torch.cuda.is_available() else "cpu"
tokenizer = AutoTokenizer.from_pretrained("dslim/bert-base-NER")
model = AutoModelForTokenClassification.from_pretrained(
"dslim/bert-base-NER")
summarizer = pipeline(
"summarization", model="facebook/bart-large-cnn", device=device)
qa = pipeline("question-answering",
model="deepset/roberta-base-squad2", device=device)
def extract_text(image):
"""
Extracts text from an image using OCR.
Args:
image (PIL.Image.Image): Input image.
Returns:
dict: Extracted text with confidence and coordinates.
Raises:
ValueError: If the input image is not a PIL Image object.
"""
result = pytesseract.image_to_data(image, output_type='dict')
n_boxes = len(result['level'])
data = {}
k = 0
for i in range(n_boxes):
if result['conf'][i] >= 0.3 and result['text'][i] != '' and result['conf'][i] != -1:
data[k] = {}
(x, y, w, h) = (result['left'][i], result['top']
[i], result['width'][i], result['height'][i])
data[k]["coordinates"] = (x, y, w, h)
text, conf = result['text'][k], result['conf'][k]
data[k]["text"] = text
data[k]["conf"] = conf
k += 1
return data
def strong_entities(question):
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
ner_results = nlp(question)
search_terms = []
current_term = ""
for token in ner_results:
if token["score"] >= 0.99:
current_term += " " + token["word"]
else:
if current_term:
search_terms.append(current_term.strip())
current_term = ""
search_terms.append(token["word"])
if current_term:
search_terms.append(current_term.strip())
print(search_terms[0].split())
return search_terms[0].split()
def wiki_search(question):
search_terms = strong_entities(question)
URL = "https://en.wikipedia.org/w/api.php"
corpus = []
for term in set(search_terms): # Removing duplicates
SEARCHPAGE = term
params = {
"action": "query",
"format": "json",
"titles": SEARCHPAGE,
"prop": "extracts",
"explaintext": True
}
response = requests.get(URL, params=params)
try:
if response.status_code == 200:
data = response.json()
for page_id, page_data in t.tqdm(data["query"]["pages"].items()):
if "extract" in page_data: # Check if extract exists
corpus.append(page_data["extract"])
else:
print("Failed to retrieve data:", response.status_code)
except Exception as e:
print("Failed to retrieve data:", e)
final_corpus = []
for text in corpus:
sections = re.split("\n\n\n== |==\n\n", text)
for section in sections:
if len(section.split()) >= 5:
final_corpus.append(section)
return " ".join(final_corpus[0:1])
def semantic_search(corpus, question):
model = SentenceTransformer("all-MiniLM-L6-v2")
question_embedding = model.encode(question)
max_similarity = -1
most_similar_doc = None
print(type(corpus[0]))
print(corpus)
for doc in t.tqdm(corpus):
if len(doc.split()) >= 130:
doc_summary = summarizer(
doc, max_length=130, min_length=30, do_sample=False)
if len(doc_summary) > 0 and "summary_text" in doc_summary[0]:
summarized_doc = doc_summary[0]["summary_text"]
else:
summarized_doc = doc
else:
summarized_doc = doc
doc_embedding = model.encode(summarized_doc)
similarity = cosine_similarity(
[question_embedding], [doc_embedding])[0][0]
if similarity > max_similarity:
max_similarity = similarity
most_similar_doc = summarized_doc
return most_similar_doc, similarity
def dm(q, a, corpus, new_q, max_history_size=5):
history = deque(maxlen=max_history_size)
history.append({"question": q, "answer": a, "corpus": corpus})
best_corpus_index = None
max_similarity = -1
for i in range(len(history)):
_, q_similarity = semantic_search([history[i]["corpus"]], new_q)
_, a_similarity = semantic_search(
[history[i]["corpus"]], history[i]["answer"])
similarity = max(q_similarity, a_similarity)
if similarity > max_similarity:
max_similarity = similarity
best_corpus_index = i
if best_corpus_index is not None:
return history[best_corpus_index]["corpus"]
else:
return corpus
def first_corp(data, question, botton=False):
if botton:
corpus = wiki_search(question)
texts = [data[i]["text"] for i in range(len(data))]
text = " ".join(texts)
corpus = [cp + " " + text for cp in corpus]
else:
texts = [data[i]["text"] for i in range(len(data))]
text = " ".join(texts)
corpus = [text]
return " ".join(corpus)
def Qa(image, new_q, internet_access=False):
old_q = ["how are you?"]
old_a = ["I am fine, thank you."]
im_text = extract_text(image)
if im_text: # Check if text is extracted
old_corpus = [first_corp(im_text, old_q[-1], botton=internet_access)]
else:
old_corpus = None
if internet_access:
if not old_corpus:
# Pass None as corpus to trigger internet access
corpus = dm(old_q[-1], old_a[-1], None, new_q)
else:
# Pass old_corpus for internet access
corpus = dm(old_q[-1], old_a[-1], old_corpus, new_q)
else:
corpus = old_corpus[0] if old_corpus else None
a = qa(question=new_q, context=corpus)
old_q.append(new_q)
old_a.append(a["answer"])
old_corpus.append(corpus)
old_conversations = "\n".join(
f"Q: {q}\nA: {a}" for q, a in zip(old_q, old_a))
return a["answer"], old_conversations
|