diff --git a/transformer_multi_input.py b/transformer_multi_input.py
new file mode 100644
index 0000000..a54ea24
--- /dev/null
+++ b/transformer_multi_input.py
@@ -0,0 +1,385 @@
+# %%
+# 生成词嵌入文件
+from tensorflow.keras import layers
+from tensorflow import keras
+from tqdm import tqdm
+import numpy as np
+import pandas as pd
+from tensorflow.keras.callbacks import ModelCheckpoint, LearningRateScheduler
+from gensim.models import Word2Vec, KeyedVectors
+from tensorflow.keras.layers import Input, LSTM, Embedding, Dense, Dropout, concatenate, Bidirectional
+from tensorflow.keras.models import Model, Sequential
+import tensorflow as tf
+from tensorflow.keras.preprocessing.sequence import pad_sequences
+from tensorflow.keras.preprocessing.text import Tokenizer
+from mymail import mail
+import os
+from tensorflow.keras.utils import to_categorical
+os.environ["CUDA_VISIBLE_DEVICES"] = "0"
+
+# %%
+
+
+class MultiHeadSelfAttention(layers.Layer):
+    def __init__(self, embed_dim, num_heads=8):
+        super(MultiHeadSelfAttention, self).__init__()
+        self.embed_dim = embed_dim
+        self.num_heads = num_heads
+        if embed_dim % num_heads != 0:
+            raise ValueError(
+                f"embedding dimension = {embed_dim} should be divisible by number of heads = {num_heads}"
+            )
+        self.projection_dim = embed_dim // num_heads
+        self.query_dense = layers.Dense(embed_dim)
+        self.key_dense = layers.Dense(embed_dim)
+        self.value_dense = layers.Dense(embed_dim)
+        self.combine_heads = layers.Dense(embed_dim)
+
+    def attention(self, query, key, value):
+        score = tf.matmul(query, key, transpose_b=True)
+        dim_key = tf.cast(tf.shape(key)[-1], tf.float32)
+        scaled_score = score / tf.math.sqrt(dim_key)
+        weights = tf.nn.softmax(scaled_score, axis=-1)
+        output = tf.matmul(weights, value)
+        return output, weights
+
+    def separate_heads(self, x, batch_size):
+        x = tf.reshape(
+            x, (batch_size, -1, self.num_heads, self.projection_dim))
+        return tf.transpose(x, perm=[0, 2, 1, 3])
+
+    def call(self, inputs):
+        # x.shape = [batch_size, seq_len, embedding_dim]
+        batch_size = tf.shape(inputs)[0]
+        query = self.query_dense(inputs)  # (batch_size, seq_len, embed_dim)
+        key = self.key_dense(inputs)  # (batch_size, seq_len, embed_dim)
+        value = self.value_dense(inputs)  # (batch_size, seq_len, embed_dim)
+        query = self.separate_heads(
+            query, batch_size
+        )  # (batch_size, num_heads, seq_len, projection_dim)
+        key = self.separate_heads(
+            key, batch_size
+        )  # (batch_size, num_heads, seq_len, projection_dim)
+        value = self.separate_heads(
+            value, batch_size
+        )  # (batch_size, num_heads, seq_len, projection_dim)
+        attention, weights = self.attention(query, key, value)
+        attention = tf.transpose(
+            attention, perm=[0, 2, 1, 3]
+        )  # (batch_size, seq_len, num_heads, projection_dim)
+        concat_attention = tf.reshape(
+            attention, (batch_size, -1, self.embed_dim)
+        )  # (batch_size, seq_len, embed_dim)
+        output = self.combine_heads(
+            concat_attention
+        )  # (batch_size, seq_len, embed_dim)
+        return output
+
+
+# %%
+class TransformerBlock(layers.Layer):
+    def __init__(self, embed_dim, num_heads, ff_dim, rate=0.1):
+        super(TransformerBlock, self).__init__()
+        self.att = MultiHeadSelfAttention(embed_dim, num_heads)
+        self.ffn = keras.Sequential(
+            [layers.Dense(ff_dim, activation="relu"),
+             layers.Dense(embed_dim), ]
+        )
+        self.layernorm1 = layers.LayerNormalization(epsilon=1e-6)
+        self.layernorm2 = layers.LayerNormalization(epsilon=1e-6)
+        self.dropout1 = layers.Dropout(rate)
+        self.dropout2 = layers.Dropout(rate)
+
+    def call(self, inputs, training):
+        attn_output = self.att(inputs)
+        attn_output = self.dropout1(attn_output, training=training)
+        out1 = self.layernorm1(inputs + attn_output)
+        ffn_output = self.ffn(out1)
+        ffn_output = self.dropout2(ffn_output, training=training)
+        return self.layernorm2(out1 + ffn_output)
+
+# %%
+
+
+class TokenAndPositionEmbedding(layers.Layer):
+    def __init__(self, maxlen, vocab_size, emded_dim):
+        super(TokenAndPositionEmbedding, self).__init__()
+        self.token_emb = layers.Embedding(
+            input_dim=vocab_size, output_dim=emded_dim)
+        self.pos_emb = layers.Embedding(input_dim=maxlen, output_dim=emded_dim)
+
+    def call(self, x):
+        maxlen = tf.shape(x)[-1]
+        positions = tf.range(start=0, limit=maxlen, delta=1)
+        positions = self.pos_emb(positions)
+        x = self.token_emb(x)
+        return x + positions
+# f = open('word2vec/userid_creative_ids.txt')
+# LEN_creative_id = -1
+# for line in f:
+#     current_line_len = len(line.strip().split(' '))
+#     LEN_creative_id = max(LEN_creative_id, current_line_len)
+# f.close()
+
+
+# %%
+def get_train_val():
+
+    # 获取 creative_id 特征
+    # f = open('tmp/userid_creative_ids.txt')
+    f = open('word2vec/userid_creative_ids.txt')
+    tokenizer = Tokenizer(num_words=NUM_creative_id)
+    tokenizer.fit_on_texts(f)
+    f.close()
+    creative_id_seq = []
+    with open('word2vec/userid_creative_ids.txt') as f:
+        for text in f:
+            creative_id_seq.append(text.strip())
+
+    sequences = tokenizer.texts_to_sequences(creative_id_seq[:900000//1])
+    X1_train = pad_sequences(
+        sequences, maxlen=LEN_creative_id, padding='post')
+
+    # 获取creative_id embedding
+    def get_creative_id_emb():
+        path = "word2vec/wordvectors_creative_id.kv"
+        wv = KeyedVectors.load(path, mmap='r')
+        creative_id_tokens = list(wv.vocab.keys())
+        embedding_dim = 128
+        embedding_matrix = np.random.randn(
+            len(creative_id_tokens)+1, embedding_dim)
+        for creative_id in creative_id_tokens:
+            embedding_vector = wv[creative_id]
+            if embedding_vector is not None:
+                index = tokenizer.texts_to_sequences([creative_id])[0][0]
+                embedding_matrix[index] = embedding_vector
+        return embedding_matrix
+
+    creative_id_emb = get_creative_id_emb()
+
+    # 获取 ad_id 特征
+    f = open('word2vec/userid_ad_ids.txt')
+    tokenizer = Tokenizer(num_words=NUM_ad_id)
+    tokenizer.fit_on_texts(f)
+    f.close()
+    ad_id_seq = []
+    with open('word2vec/userid_ad_ids.txt') as f:
+        for text in f:
+            ad_id_seq.append(text.strip())
+
+    sequences = tokenizer.texts_to_sequences(ad_id_seq[:900000//1])
+    X2_train = pad_sequences(
+        sequences, maxlen=LEN_ad_id, padding='post')
+
+    def get_ad_id_emb():
+        path = "word2vec/wordvectors_ad_id.kv"
+        wv = KeyedVectors.load(path, mmap='r')
+        ad_id_tokens = list(wv.vocab.keys())
+        embedding_dim = 128
+        embedding_matrix = np.random.randn(
+            len(ad_id_tokens)+1, embedding_dim)
+        for ad_id in ad_id_tokens:
+            embedding_vector = wv[ad_id]
+            if embedding_vector is not None:
+                index = tokenizer.texts_to_sequences([ad_id])[0][0]
+                embedding_matrix[index] = embedding_vector
+        return embedding_matrix
+
+    ad_id_emb = get_ad_id_emb()
+
+    # 获取 product_id 特征
+    # f = open('tmp/userid_product_ids.txt')
+    f = open('word2vec/userid_product_ids.txt')
+    tokenizer = Tokenizer(num_words=NUM_product_id)
+    tokenizer.fit_on_texts(f)
+    f.close()
+    product_id_seq = []
+    with open('word2vec/userid_product_ids.txt') as f:
+        for text in f:
+            product_id_seq.append(text.strip())
+
+    sequences = tokenizer.texts_to_sequences(product_id_seq[:900000//1])
+    X3_train = pad_sequences(
+        sequences, maxlen=LEN_product_id, padding='post')
+
+    # 获取product_id embedding
+    def get_product_id_emb():
+        path = "word2vec/wordvectors_product_id.kv"
+        wv = KeyedVectors.load(path, mmap='r')
+        product_id_tokens = list(wv.vocab.keys())
+        embedding_dim = 128
+        embedding_matrix = np.random.randn(
+            len(product_id_tokens)+1, embedding_dim)
+        for product_id in product_id_tokens:
+            embedding_vector = wv[product_id]
+            if embedding_vector is not None:
+                index = tokenizer.texts_to_sequences([product_id])[0][0]
+                embedding_matrix[index] = embedding_vector
+        return embedding_matrix
+
+    product_id_emb = get_product_id_emb()
+
+    # 获得age标签
+    user_train = pd.read_csv(
+        'data/train_preliminary/user.csv').sort_values(['user_id'], ascending=(True,))
+    Y_gender = user_train['gender'].values
+    Y_age = user_train['age'].values
+    Y_gender = Y_gender - 1
+    Y_age = Y_age - 1
+    Y_age = to_categorical(Y_age)
+    num_examples = Y_age.shape[0]
+    train_examples = int(num_examples * 0.9)
+
+    # 分别对应 x1_train x1_val x2_train x2_val y_train y_val
+    return X1_train[:train_examples], X1_train[train_examples:], X2_train[:train_examples], X2_train[train_examples:], X3_train[:train_examples], X3_train[train_examples:], Y_age[:train_examples], Y_age[train_examples:], creative_id_emb, ad_id_emb, product_id_emb
+
+
+# %%
+NUM_creative_id = 2481135+1
+NUM_ad_id = 2264190+1
+NUM_product_id = 33273+1
+
+vocab_size = NUM_creative_id 
+maxlen = LEN_creative_id
+
+
+def get_age_model(creative_id_emb, ad_id_emb, product_id_emb):
+    embed_dim = 128  # Embedding size for each token
+    num_heads = 2  # Number of attention heads
+    ff_dim = 32  # Hidden layer size in feed forward network inside transformer
+    # shape：(sequence长度, )
+    # first input
+    input_creative_id = Input(shape=(None,), name='creative_id')
+    # x1 = Embedding(input_dim=NUM_creative_id,
+    #                output_dim=128,
+    #                weights=[creative_id_emb],
+    #                trainable=True,
+    #                input_length=LEN_creative_id,
+    #                mask_zero=True)(input_creative_id)
+    x1 = TokenAndPositionEmbedding(
+        maxlen, vocab_size, embed_dim)(input_creative_id)
+    x1 = TransformerBlock(embed_dim, num_heads, ff_dim)(x1)
+    x1 = GlobalAveragePooling1D()(x1)
+    x1 = layers.Dropout(0.1)(x1)
+    x1 = layers.Dense(20, activation="relu")(x1)
+    x1 = layers.Dropout(0.1)(x1)
+    outputs = layers.Dense(10, activation="softmax")(x1)
+    # x1 = LSTM(1024, return_sequences=True)(x1)
+    # x1 = LSTM(512, return_sequences=True)(x1)
+    # x1 = LSTM(256, return_sequences=False)(x1)
+
+    # second input
+    # input_ad_id = Input(shape=(None,), name='ad_id')
+    # x2 = Embedding(input_dim=NUM_ad_id,
+    #                output_dim=128,
+    #                weights=[ad_id_emb],
+    #                trainable=True,
+    #                input_length=LEN_ad_id,
+    #                mask_zero=True)(input_ad_id)
+    # x2 = LSTM(1024, return_sequences=True)(x2)
+    # x2 = LSTM(512, return_sequences=True)(x2)
+    # x2 = LSTM(256, return_sequences=False)(x2)
+
+    # third input
+    # input_product_id = Input(shape=(None,), name='product_id')
+    # x3 = Embedding(input_dim=NUM_product_id,
+    #                output_dim=128,
+    #                weights=[product_id_emb],
+    #                trainable=True,
+    #                input_length=LEN_product_id,
+    #                mask_zero=True)(input_product_id)
+    # x3 = LSTM(1024, return_sequences=True)(x3)
+    # x3 = LSTM(512, return_sequences=True)(x3)
+    # x3 = LSTM(256, return_sequences=False)(x3)
+
+    # concat x1 x2
+    # x = concatenate([x1, x2, x3])
+    # x = x1 + x2 + x3
+    # x = Dense(128)(x)
+    # x = Dropout(0.1)(x)
+    # output_y = Dense(10, activation='softmax')(x)
+
+    # model = Model([input_creative_id, input_ad_id, input_product_id], output_y)
+    model = Model(input_creative_id, outputs)
+    model.compile(loss='categorical_crossentropy',
+                  optimizer='adam', metrics=['accuracy'])
+    model.summary()
+
+    return model
+
+
+# %%
+mail('start getting train data')
+x1_train, x1_val, x2_train, x2_val, x3_train, x3_val, y_train, y_val, creative_id_emb, ad_id_emb, product_id_emb = get_train_val()
+mail('get train data done.')
+
+model = get_age_model(creative_id_emb, ad_id_emb, product_id_emb)
+# %%
+# %%
+# 测试数据格式(batch_size, sequence长度)
+# x1 = np.array([1, 2, 3, 4]).reshape(1, -1)
+# x2 = np.array([1, 2, 3, 4]).reshape(1, -1)
+# model.predict([x1, x2])
+
+
+# %%
+checkpoint = ModelCheckpoint("tmp/age_epoch_{epoch:02d}.hdf5", monitor='val_loss', verbose=1,
+                             save_best_only=False, mode='auto', period=1)
+# %%
+# model.fit(
+#     {'creative_id': x1_train, 'ad_id': x2_train},
+#     y_train,
+#     validation_data=([x1_val, x2_val], y_val),
+#     epochs=5,
+#     batch_size=256,
+#     callbacks=[checkpoint],
+# )
+
+# %%
+try:
+    mail('start train lstm')
+    model.fit(
+        {'creative_id': x1_train, 'ad_id': x2_train, 'product_id': x3_train},
+        y_train,
+        validation_data=([x1_val, x2_val, x3_val], y_val),
+        epochs=3,
+        batch_size=256,
+        callbacks=[checkpoint],
+    )
+    mail('train lstm done!!!')
+except Exception as e:
+    e = str(e)
+    mail('train lstm failed!!! ' + e)
+
+
+# %%
+# model.load_weights('tmp\gender_epoch_01.hdf5')
+
+
+# # %%
+# if debug:
+#     sequences = tokenizer.texts_to_sequences(
+#         creative_id_seq[900000:])
+# else:
+#     sequences = tokenizer.texts_to_sequences(
+#         creative_id_seq[900000:])
+
+# X_test = pad_sequences(sequences, maxlen=LEN_creative_id)
+# # %%
+# y_pred = model.predict(X_test, batch_size=4096)
+
+# y_pred = np.where(y_pred > 0.5, 1, 0)
+# y_pred = y_pred.flatten()
+
+# # %%
+# y_pred = y_pred+1
+# # %%
+# res = pd.DataFrame({'predicted_gender': y_pred})
+# res.to_csv(
+#     'data/ans/lstm_gender.csv', header=True, columns=['predicted_gender'], index=False)
+
+
+# # %%
+# mail('predict lstm gender done')
+
+# %%