app.py

import transformers

import gradio as gr
import tensorflow as tf

from official.nlp import optimization  # to create AdamW optimizer

MODEL_DIRECTORY = 'save/modelV1'
PRETRAINED_MODEL_NAME = 'dbmdz/bert-base-german-cased'
TOKENIZER = transformers.BertTokenizer.from_pretrained(PRETRAINED_MODEL_NAME)
MAX_SEQUENCE_LENGTH = 256
EPOCHS = 2
OPTIMIZER = 'adamw'
INIT_LR = 3e-5
LOSS = tf.keras.losses.BinaryCrossentropy(from_logits=False)
METRICS = tf.metrics.BinaryAccuracy()

def compile_model(model):
    steps_per_epoch = 10
    num_train_steps = steps_per_epoch * EPOCHS
    num_warmup_steps = int(0.1*num_train_steps)

    optimizer = optimization.create_optimizer(
        init_lr=INIT_LR,
        num_train_steps=steps_per_epoch,
        num_warmup_steps=num_warmup_steps,
        optimizer_type=OPTIMIZER
    )

    model.compile(optimizer=optimizer, loss=LOSS, metrics=[METRICS])
    return model

hs_detection_model = tf.keras.models.load_model(MODEL_DIRECTORY, compile=False)  #tf.keras.models.load_model('save/kerasmodel/model.h5') #tf.saved_model.load('save/model') #tf.keras.models.load_model('save/model')
compile_model(hs_detection_model)

def encode(sentences):
    return TOKENIZER.batch_encode_plus(
        sentences,
        max_length=MAX_SEQUENCE_LENGTH, # set the length of the sequences
        add_special_tokens=True, # add [CLS] and [SEP] tokens
        return_attention_mask=True,
        return_token_type_ids=False, # not needed for this type of ML task
        pad_to_max_length=True, # add 0 pad tokens to the sequences less than max_length
        return_tensors='tf'
    )

def inference(sentence):
    print(sentence)
    encoded_sentence = encode([sentence])
    print(encoded_sentence)
    predicition = hs_detection_model.predict(encoded_sentence.values())
    print(predicition)
    return predicition

iface = gr.Interface(fn=inference, inputs="text", outputs="text") #, live=True)
iface.launch()