@myh1234567
Could you please verify that all input data (x, y, etc.) passed to your model is of the correct type (tf.float32, tf.float64, etc.) as expected by the model's architecture and the loss function ? Also, in order to expedite the trouble-shooting process, please provide a code snippet to reproduce the issue reported here. Thank you!

@myh1234567 Could you please verify that all input data (x, y, etc.) passed to your model is of the correct type (tf.float32, tf.float64, etc.) as expected by the model's architecture and the loss function ? Also, in order to expedite the trouble-shooting process, please provide a code snippet to reproduce the issue reported here. Thank you!

the following is the code snippent,

import tensorflow as tf
import numpy as np
import scipy.sparse as sp
import keras
from keras.utils import to_categorical

physical_device = tf.config.list_physical_devices("GPU")
print("===================================== num gpu:", len(physical_device))

def convert_sparse_matrix_to_sparse_tensor(X) -> tf.SparseTensor:
""" Convert a scipy sparse matrix to a SparseTensor."""
coo = X.tocoo()
indices = np.mat([coo.row, coo.col]).transpose()
return tf.sparse.SparseTensor(indices, coo.data, coo.shape)

X_transformed_train_db = sp.rand(100, 1000, density=0.1, format='coo')
X_train_db = convert_sparse_matrix_to_sparse_tensor(X_transformed_train_db)
X_train_db = tf.sparse.reorder(X_train_db)
y_train_db = np.random.randint(0, 3, size=(100,)) # Let's assume 3 classes
y_train_db_ = to_categorical(y_train_db)

gpus = tf.config.list_logical_devices("GPU")
strategy = tf.distribute.MirroredStrategy(gpus)

with strategy.scope():
opti = keras.optimizers.Adam(learning_rate=0.0001)
input_dimension = X_transformed_train_db.shape[1]
model_db = keras.Sequential()
model_db.add(keras.layers.Dense(1000, kernel_initializer=keras.initializers.HeNormal(seed=1), activation='relu', input_dim=input_dimension))
model_db.add(keras.layers.Dropout(0.1))
model_db.add(keras.layers.Dense(500, kernel_initializer=keras.initializers.HeNormal(seed=2), activation='relu'))
model_db.add(keras.layers.Dropout(0.1))
model_db.add(keras.layers.Dense(200, kernel_initializer=keras.initializers.HeNormal(), activation='relu'))
model_db.add(keras.layers.Dropout(0.1))
model_db.add(keras.layers.Dense(y_train_db_.shape[1], kernel_initializer=keras.initializers.RandomNormal(mean=0.0, stddev=0.05, seed=4), activation='softmax'))
model_db.compile(optimizer=opti, loss='categorical_crossentropy',metrics=['accuracy'])

history_db = model_db.fit(X_transformed_train_db, y_train_db_, epochs=3)

tensorflow 2.16.1
keras 3.3.3
scipy 1.13.1
numpy 1.26.4

thank you