WebraftAI/Text-Completion

1import streamlit as st2import tensorflow as tf3from keras.layers import Input, Dense, Embedding, MultiHeadAttention4from keras.layers import Dropout, LayerNormalization5from keras.models import Model6from keras.utils import pad_sequences7from keras import layers8import numpy as np9import logging10logging.getLogger('tensorflow').setLevel(logging.ERROR)11class TransformerChatbot(Model):12    def __init__(self, vocab_size, max_len, d_model, n_head, ff_dim, dropout_rate):13        super(TransformerChatbot, self).__init__()14        self.embedding = Embedding(vocab_size, d_model)15        self.attention = MultiHeadAttention(num_heads=n_head, key_dim=d_model)16        self.norm1 = LayerNormalization(epsilon=1e-6)17        self.dropout1 = Dropout(dropout_rate)18        self.dense1 = Dense(ff_dim, activation="relu")19        self.dense2 = Dense(d_model)20        self.norm2 = LayerNormalization(epsilon=1e-6)21        self.dropout2 = Dropout(dropout_rate)22        self.flatten = tf.keras.layers.Flatten()23        self.fc = Dense(vocab_size, activation="softmax")24        self.max_len = max_len25 26    def call(self, inputs):27        x = self.embedding(inputs)28        # Masking29        mask = self.create_padding_mask(inputs)30        attn_output = self.attention(x, x, x, attention_mask=mask)31        x = x + attn_output32        x = self.norm1(x)33        x = self.dropout1(x)34        x = self.dense1(x)35        x = self.dense2(x)36        x = self.norm2(x)37        x = self.dropout2(x)38        x = self.fc(x)39        return x40 41    def create_padding_mask(self, seq):42        mask = tf.cast(tf.math.equal(seq, 0), tf.float32)43        return mask[:, tf.newaxis, tf.newaxis, :]44st.title("UniGLM TEXT completion Model")45st.subheader("Next Word Prediction AI Model by Webraft-AI")46#Picking what NLP task you want to do47option = st.selectbox('Model',('13M','26M')) #option is stored in this variable48#Textbox for text user is entering49st.subheader("Enter a word from which a sentence / word would be predicted")50text2 = st.text_input('Enter word: ') #text is stored in this variable51 52if option == '13M':53    with open("data2.txt","r") as f:54        text = f.read()55    text = text.lower()56    words = text.split()57    loaded_dict = np.load("dict_predict3.bin.npz", allow_pickle=True)58    word_to_num = loaded_dict["word_to_num"].item()59    num_to_word = loaded_dict["num_to_word"].item()60    X = []61    Y = []62 63    for i in range(len(words)-1):64        word = words[i]65        next_word = words[i+1]66        X.append(word_to_num[word])67        Y.append(word_to_num[next_word])68    Y.append(0)69 70    X.append(word_to_num[words[-1]])71    72    X_train = pad_sequences([X])73    y_train = pad_sequences([Y])74    vocab_size = 10000075    max_len = 176    d_model = 64  # 64 , 102477    n_head = 4  # 8 , 1678    ff_dim = 256  # 256 , 204879    dropout_rate = 0.1  # 0.5 , 0.280 81 82    chatbot = TransformerChatbot(vocab_size, max_len, d_model, n_head, ff_dim, dropout_rate)83    chatbot.load_weights("predict3")84    chatbot.build(input_shape=(None, max_len)) # Build the model85    chatbot.compile(optimizer="adam", loss="sparse_categorical_crossentropy")86    87    for i in range(1):88        other_text1 = text289        other_text1 = other_text1.lower()90        other_words1 = other_text1.split()91        other_num1 = [word_to_num[word] for word in other_words1]92        given_X1 = other_num193        input_sequence1 = pad_sequences([given_X1], maxlen=max_len, padding='post')94        output_sentence = other_text1+""95        for _ in range(13):96            predicted_token = np.argmax(chatbot.predict(input_sequence1), axis=-1)97            predicted_token = predicted_token.item()98            out = num_to_word[predicted_token]99            100    101            output_sentence += " " + out102            if out == ".":103                break104            given_X1 = given_X1[1:]105            given_X1.append(predicted_token)106            input_sequence1 = pad_sequences([given_X1], maxlen=max_len, padding='post')107 108        out2 = output_sentence109        110    111else:112    with open("data2.txt","r") as f:113        text = f.read()114    text = text.lower()115    words = text.split()116    loaded_dict = np.load("dict_predict1.bin.npz", allow_pickle=True)117    word_to_num = loaded_dict["word_to_num"].item()118    num_to_word = loaded_dict["num_to_word"].item()119    X = []120    Y = []121    for i in range(len(words)-1):122        word = words[i]123        next_word = words[i+1]124        X.append(word_to_num[word])125        Y.append(word_to_num[next_word])126    Y.append(0)127 128    X.append(word_to_num[words[-1]])129    X_train = pad_sequences([X])130    y_train = pad_sequences([Y])131    vocab_size = 100000132    max_len = 1133    d_model = 128  # 64 , 1024134    n_head = 4  # 8 , 16135    ff_dim = 256  # 256 , 2048136    dropout_rate = 0.1  # 0.5 , 0.2137 138 139    chatbot = TransformerChatbot(vocab_size, max_len, d_model, n_head, ff_dim, dropout_rate)140    chatbot.load_weights("predict1")141    chatbot.build(input_shape=(None, max_len)) # Build the model142    chatbot.compile(optimizer="adam", loss="sparse_categorical_crossentropy")143    144    for i in range(1):145        other_text1 = text2146        other_text1 = other_text1.lower()147        other_words1 = other_text1.split()148        other_num1 = [word_to_num[word] for word in other_words1]149        given_X1 = other_num1150        input_sequence1 = pad_sequences([given_X1], maxlen=max_len, padding='post')151        output_sentence = other_text1+""152        for _ in range(10):153            predicted_token = np.argmax(chatbot.predict(input_sequence1), axis=-1)154            predicted_token = predicted_token.item()155            out = num_to_word[predicted_token]156            157    158            output_sentence += " " + out159            if out == ".":160                break161            given_X1 = given_X1[1:]162            given_X1.append(predicted_token)163            input_sequence1 = pad_sequences([given_X1], maxlen=max_len, padding='post')164 165        out2 = output_sentence166 167 168 169 170 171    172st.write("Predicted Text: ")173st.write(out2)