File size: 4,900 Bytes
df92393
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import argparse
import numpy as np
import connexion
from flask_cors import CORS
from flask import render_template, redirect, send_from_directory

import utils.path_fixes as pf
from utils.f import ifnone

from model_api import get_details

app = connexion.FlaskApp(__name__, static_folder="client/dist", specification_dir=".")
flask_app = app.app
CORS(flask_app)

parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument("--debug", action="store_true", help=" Debug mode")
parser.add_argument("--port", default=5051, help="Port to run the app. ")

# Flask main routes
@app.route("/")
def hello_world():
    return redirect("client/exBERT.html")

# send everything from client as static content
@app.route("/client/<path:path>")
def send_static_client(path):
    """ serves all files from ./client/ to ``/client/<path:path>``
    :param path: path from api call
    """
    return send_from_directory(str(pf.CLIENT_DIST), path)

# ======================================================================
## CONNEXION API ##
# ======================================================================
def get_model_details(**request):
    """Get important information about a model, like the number of layers and heads
    
    Args:
        request['model']: The model name
    Returns:
        {
            status: 200,
            payload: {
                nlayers (int)
                nheads (int)
            }
        }
    """
    mname = request['model']
    deets = get_details(mname)

    info = deets.config
    nlayers = info.num_hidden_layers
    nheads = info.num_attention_heads

    payload_out = {
        "nlayers": nlayers,
        "nheads": nheads,
    }

    return {
        "status": 200,
        "payload": payload_out,
    }

def get_attentions_and_preds(**request):
    """For a sentence, at a layer, get the attentions and predictions
    
    Args:
        request['model']: Model name
        request['sentence']: Sentence to get the attentions for
        request['layer']: Which layer to extract from
    Returns:
        {
            status: 200
            payload: {
                aa: {
                    att: Array((nheads, ntoks, ntoks))
                    left: [{
                        text (str), 
                        topk_words (List[str]),
                        topk_probs (List[float])
                    }, ...]
                    right: [{
                        text (str), 
                        topk_words (List[str]),
                        topk_probs (List[float])
                    }, ...]
                }
            }
        }
    """
    model = request["model"]
    details = get_details(model)

    sentence = request["sentence"]
    layer = int(request["layer"])

    deets = details.from_sentence(sentence)

    payload_out = deets.to_json(layer)

    return {
        "status": 200,
        "payload": payload_out
    }

def update_masked_attention(**request):
    """From tokens and indices of what should be masked, get the attentions and predictions
    
    payload = request['payload']
    Args:
        payload['model'] (str): Model name
        payload['tokens'] (List[str]): Tokens to pass through the model
        payload['sentence'] (str): Original sentence the tokens came from
        payload['mask'] (List[int]): Which indices to mask
        payload['layer'] (int): Which layer to extract information from
    Returns:
        {
            status: 200
            payload: {
                aa: {
                    att: Array((nheads, ntoks, ntoks))
                    left: [{
                        text (str), 
                        topk_words (List[str]),
                        topk_probs (List[float])
                    }, ...]
                    right: [{
                        text (str), 
                        topk_words (List[str]),
                        topk_probs (List[float])
                    }, ...]
                }
            }
        }
    """
    payload = request["payload"]

    model = payload['model']
    details = get_details(model)

    tokens = payload["tokens"]
    sentence = payload["sentence"]
    mask = payload["mask"]
    layer = int(payload["layer"])

    MASK = details.tok.mask_token
    mask_tokens = lambda toks, maskinds: [
        t if i not in maskinds else ifnone(MASK, t) for (i, t) in enumerate(toks)
    ]

    token_inputs = mask_tokens(tokens, mask)

    deets = details.from_tokens(token_inputs, sentence)
    payload_out = deets.to_json(layer)

    return {
        "status": 200,
        "payload": payload_out,
    }

app.add_api("swagger.yaml")

# Setup code
if __name__ != "__main__":
    print("SETTING UP ENDPOINTS")

# Then deploy app
else:
    args, _ = parser.parse_known_args()
    print("Initiating app")
    app.run(port=args.port, use_reloader=False, debug=args.debug)