File size: 6,475 Bytes
a5187e1
 
 
2abd8dc
 
a5187e1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d88718a
a5187e1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d88718a
a5187e1
286646b
 
a5187e1
 
 
 
 
 
 
 
 
 
 
 
 
d36accc
0341a18
d36accc
a5187e1
0904a51
a5187e1
 
 
 
 
 
 
a886c12
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

from transformers import AutoTokenizer

# from huggingface_hub import notebook_login
# notebook_login()

from transformers import AutoTokenizer
tokenizer1 = AutoTokenizer.from_pretrained("Reham721/Subjective_QG")
tokenizer2 = AutoTokenizer.from_pretrained("Reham721/MCQs")

from transformers import AutoModelForSeq2SeqLM

model1 = AutoModelForSeq2SeqLM.from_pretrained("Reham721/Subjective_QG")
model2 = AutoModelForSeq2SeqLM.from_pretrained("Reham721/MCQs")


from arabert.preprocess import ArabertPreprocessor
from transformers import pipeline

prep = ArabertPreprocessor("aubmindlab/araelectra-base-discriminator") #or empty string it's the same
qa_pipe =pipeline("question-answering",model="wissamantoun/araelectra-base-artydiqa")

def generate_questions(model,tokenizer, input_sequence):

    # Tokenize input sequence
    input_ids = tokenizer.encode(input_sequence, return_tensors='pt')

    # Generate questions
    outputs = model.generate(
    input_ids=input_ids,
    max_length=200, # Set a shorter maximum length for shorter questions
    num_beams=3, # Use fewer beams for faster generation and to avoid overfitting
    no_repeat_ngram_size=3, # Allow some repetition to avoid overly generic questions
    early_stopping=True, # Stop generation when all beams are finished
    temperature=1, # Use a lower temperature for more conservative questions
    num_return_sequences=3, # Generate more questions per input
)

    # Decode questions
    questions = []
    for output in outputs:
        output_text = tokenizer.decode(output, skip_special_tokens=True)
        questions.append(output_text)

    return questions

def get_sorted_questions(questions, context):
    dic = {}
    context = prep.preprocess(context)
    for question in questions:
        print(question)
        result = qa_pipe(question=question,context=context)
        print(result)
        dic.update({question: result["score"]})

    return dict(sorted(dic.items(), key=lambda item: item[1], reverse=True))



import unicodedata
import arabic_reshaper
from bidi.algorithm import get_display

def is_arabic(text):
    # Reshape the text using the arabic_reshaper library
    reshaped_text = arabic_reshaper.reshape(text)
    # Determine the directionality of the text using the get_display() function from the bidi algorithm library
    bidi_text = get_display(reshaped_text)
    # Check if the text contains any non-Arabic letters
    for char in bidi_text:
        if char.isalpha() and unicodedata.name(char).startswith('ARABIC') == False:
            return False
    return True

import random
import re
def generate_distractors(question, answer, context, num_distractors=3, k=10):
    input_sequence = f'{question} <sep> {answer} <sep> {context}'
    input_ids = tokenizer2.encode(input_sequence, return_tensors='pt')

    # Generate distractors using model.generate()
    outputs = model2.generate(
        input_ids,
        do_sample=True, max_length=50, top_k=50, top_p=0.95, num_return_sequences=num_distractors, no_repeat_ngram_size=2)

    # Convert outputs to list of strings
    distractors = []
    for output in outputs:
        decoded_output = tokenizer2.decode(output, skip_special_tokens=True)
        distractor_elements = [re.sub(r'<[^>]*>', '', element.strip()) for element in re.split(r'(<[^>]*>)|(?:None)', decoded_output) if element]
        distractor_elements = [element for element in distractor_elements if element]
        distractor_elements = [element for element in distractor_elements if is_arabic(element)]
        distractors.append(distractor_elements)
    distractors = [element for sublist in distractors for element in sublist]


    # Remove duplicate distractors
    unique_distractors = []
    for distractor in distractors:
        if distractor not in unique_distractors and distractor != answer:
            unique_distractors.append(distractor)


    # If there are not enough unique distractors, generate more until there are
    while len(unique_distractors) < num_distractors:

        outputs = model2.generate(
            input_ids,
            do_sample=True,
            max_length=50,
            top_k=50,
            top_p=0.95,
            num_return_sequences=num_distractors-len(unique_distractors),
            no_repeat_ngram_size=2)
        for output in outputs:
            decoded_output = tokenizer2.decode(output, skip_special_tokens=True)
            distractor_elements = [re.sub(r'<[^>]*>', '', element.strip()) for element in re.split(r'(<[^>]*>)|(?:None)', decoded_output) if element]
            distractor_elements = [element for element in distractor_elements if element]
            distractor_elements = [element for element in distractor_elements if is_arabic(element)]
            if decoded_output not in unique_distractors and decoded_output not in unique_distractors and decoded_output != answer:
                unique_distractors.append(decoded_output)
            if len(unique_distractors) >= num_distractors:
                break

    random.shuffle(unique_distractors)

    # Select k top distractors if more than k obtained in step 2
    if len(unique_distractors) > k:
        unique_distractors = sorted(unique_distractors, key=lambda x: random.random())[:k]

    # Select num_distractors distractors
    distractor_subset = random.sample(unique_distractors, num_distractors)

    return distractor_subset



import gradio as gr

context = gr.inputs.Textbox(lines=5,placeholder="Enter paragraph/context here...")
answer = gr.inputs.Textbox(lines=3, placeholder="Enter answer/keyword here...")
question_type = gr.inputs.Radio(choices=["Subjective", "MCQ"], label="Question type")
question = gr.outputs.Textbox( type="text", label="Question")

def generate_question(context,answer,question_type):
    article = answer+"<sep>"+context
    output = generate_questions(model1, tokenizer1, article)
    result = get_sorted_questions(output, context)
    if question_type == "Subjective":
        return next(iter(result))
    else:
        mcqs = generate_distractors(question, answer, context)
        # mcqs[3] = answer
        # random.shuffle(mcqs)
        return next(iter(result))+"\n"+"-" + mcqs[0]+"\n"+"-" + mcqs[1]+"\n"+"-" + mcqs[2]+"\n" +"-" + answer +"\n"

        
iface = gr.Interface(
  fn=generate_question,
  inputs=[context,answer,question_type],
  outputs=question,
  list_outputs=True,
  rtl=True)

iface.launch(debug=True,share=False) # will create a temporary sharable link