File size: 5,007 Bytes
3411193
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2df020e
3411193
 
 
 
 
 
 
 
 
 
 
2df020e
 
 
3411193
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8619f75
2df020e
400e02d
 
 
 
3411193
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import streamlit as st
from transformers import pipeline
from transformers.tokenization_utils import TruncationStrategy

import tokenizers
import pandas as pd
import requests

st.set_page_config(
     page_title='AlephBERT Demo',
     page_icon="🥙",
     initial_sidebar_state="expanded",
)

models = {
    "AlephBERT-base": {
        "name_or_path":"onlplab/alephbert-base",
        "description":"AlephBERT base model",
    },
    "HeBERT-base-TAU": {
        "name_or_path":"avichr/heBERT",
        "description":"HeBERT model created by TAU"
    },
    "mBERT-base-multilingual-cased": {
        "name_or_path":"bert-base-multilingual-cased",
        "description":"Multilingual BERT model"
    }
}

@st.cache(show_spinner=False)
def get_json_from_url(url):
    return models
    return requests.get(url).json()

# models = get_json_from_url('https://huggingface.co/spaces/biu-nlp/AlephBERT/raw/main/models.json')



@st.cache(show_spinner=False, hash_funcs={tokenizers.Tokenizer: str})
def load_model(model):
    pipe = pipeline('fill-mask', models[model]['name_or_path'])
    def do_tokenize(inputs):
        return pipe.tokenizer(
                inputs,
                add_special_tokens=True,
                return_tensors=pipe.framework,
                padding=True,
                truncation=TruncationStrategy.DO_NOT_TRUNCATE,
            )

    def _parse_and_tokenize(
        inputs, tokenized=False, **kwargs
    ):
        if not tokenized:
            inputs = do_tokenize(inputs)
        return inputs

    pipe._parse_and_tokenize = _parse_and_tokenize
    
    return pipe, do_tokenize





st.title('AlephBERT🥙')
st.sidebar.markdown(
    """<div><a  target="_blank" href="https://nlp.biu.ac.il/~rtsarfaty/onlp#"><img src="https://nlp.biu.ac.il/~rtsarfaty/static/landing_static/img/onlp_logo.png"  style="filter: invert(100%);display: block;margin-left: auto;margin-right: auto;
  width: 70%;"></a>
      <p style="color:white; font-size:13px; font-family:monospace; text-align: center">AlephBERT Demo &bull; <a href="https://nlp.biu.ac.il/~rtsarfaty/onlp#" style="text-decoration: none;color: white;"  target="_blank">ONLP Lab</a></p></div>
      <br>""",
    unsafe_allow_html=True,
)

mode = 'Models'

if mode == 'Models':
    model = st.sidebar.selectbox(
     'Select Model',
     list(models.keys()))
    masking_level = st.sidebar.selectbox('Masking Level:', ['Tokens', 'SubWords'])
    n_res = st.sidebar.number_input(
        'Number Of Results',
        format='%d',
        value=5,
        min_value=1,
        max_value=100)
    
    model_tags = model.split('-')
    model_tags[0] = 'Model:' + model_tags[0] 

    st.markdown(''.join([f'<span style="color:white; font-size:13px; font-family:monospace; background-color: #f63766;margin:3px;padding:8px;border-radius: 5px;">{tag}</span>' for tag in model_tags]),unsafe_allow_html=True)
    st.markdown('___')

    unmasker, tokenize = load_model(model)
            
    input_text = st.text_input('Insert text you want to mask', '')
    if input_text:
        input_masked = None
        tokenized = tokenize(input_text)
        ids = tokenized['input_ids'].tolist()[0]
        subwords = unmasker.tokenizer.convert_ids_to_tokens(ids)
        
        if masking_level == 'Tokens':
            tokens = str(input_text).split()
            mask_idx = st.selectbox('Select token to mask:', [None] + list(range(len(tokens))), format_func=lambda i: tokens[i] if i else '')
            if mask_idx is not None:
                input_masked = ' '.join(token if i != mask_idx else '[MASK]' for i, token in enumerate(tokens))
                display_input = input_masked
        if masking_level == 'SubWords':
            tokens = subwords
            idx = st.selectbox('Select token to mask:', list(range(0,len(tokens)-1)), format_func=lambda i: tokens[i] if i else '')
            tokenized['input_ids'][0][idx] = unmasker.tokenizer.mask_token_id
            ids = tokenized['input_ids'].tolist()[0]
            display_input = ' '.join(unmasker.tokenizer.convert_ids_to_tokens(ids[1:-1]))
            if idx:
                input_masked = tokenized
                
        if input_masked: 
            st.markdown('#### Input:')
            ids = tokenized['input_ids'].tolist()[0]
            subwords = unmasker.tokenizer.convert_ids_to_tokens(ids)
            st.markdown(f'<p dir="rtl">{display_input}</p>',
                        unsafe_allow_html=True,
            )
            st.markdown('#### Outputs:')
            with st.spinner(f'Running {model_tags[0]} (may take a minute)...'):
                res = unmasker(input_masked, tokenized=masking_level == 'SubWords', top_k=n_res)
                if res:
                    res = [{'Prediction':r['token_str'], 'Completed Sentence':r['sequence'].replace('[SEP]', '').replace('[CLS]', ''), 'Score':r['score']} for r in res]
                    res_table = pd.DataFrame(res)
                    st.table(res_table)