migrating keras + tensorflow cpu to tensorflow-gpu keras modelling

Question

I'm a beginner of data-science and by now, I'm trying to migrate old code keras cpu modelling to gpu-tensorflow.

fyi: I'm following instruction on prof.jeffheaton videos: https://www.youtube.com/watch?v=qrkEYf-YDyI | [2020, TensorFlow 2.0 GPU (CUDA), Keras, & Python 3.7 in Windows 10] and install environment as noticed here: https://github.com/jeffheaton/t81_558_deep_learning/blob/master/tensorflow-gpu.yml

and from nvidia, i had installed: ms.visual code 2019

• cuda_11.5.1_496.13_windows -- include driver.exe
• cudnn-windows-x86_64-8.3.1.22_cuda11.5-archive.zip

and as result on phyton: import tensorflow as tf
i got result:
2021-12-04 18:02:41.467262: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll

and for: tf.config.list_physical_devices('GPU')
it result:
2021-12-04 18:02:41.467262: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll
2021-12-04 18:04:50.986777: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1555] Found device 0 with properties:
pciBusID: 0000:01:00.0 name: NVIDIA GeForce 920MX computeCapability: 5.0 coreClock: 0.993GHz coreCount: 2 deviceMemorySize: 2.00GiB deviceMemoryBandwidth: 13.41GiB/s
2021-12-04 18:04:50.986857: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll
2021-12-04 18:04:50.986900: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_10.dll
2021-12-04 18:04:50.986939: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cufft64_10.dll
2021-12-04 18:04:50.986978: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library curand64_10.dll
2021-12-04 18:04:50.987016: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusolver64_10.dll
2021-12-04 18:04:50.987054: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusparse64_10.dll
2021-12-04 18:04:50.987093: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudnn64_7.dll
2021-12-04 18:04:50.987155: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1697] Adding visible gpu devices: 0
[PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]

---

and here the cpu modeling code:

from keras.models import Sequential
from keras.layers import Dense, LSTM

###CPU Build from
import math
from keras.models import Sequential
from keras.layers import Dense, Activation, Conv1D ,MaxPooling1D, Dropout
from keras.layers.recurrent import LSTM
from keras.utils.vis_utils import plot_model
from keras.metrics import RootMeanSquaredError as rmse
from keras import optimizers

and it still run normal for cpu code line:

model = Sequential()
model.add(Conv1D(filters=256, kernel_size=2, activation='relu',padding = 'same',input_shape=(120,1)))
model.add(MaxPooling1D(pool_size=2))
model.add(LSTM(100, return_sequences = True))
model.add(Dropout(0.3))
model.add(LSTM(100, return_sequences = False))
model.add(Dropout(0.3))
model.add(Dense(units=1, activation = 'sigmoid'))
model.compile(optimizer='adam', loss= 'mse' , metrics = [rmse()])

migrate to tensorflow gpu keras

from tensorflow import keras ##GPU
from tensorflow.keras import layers ##GPU
from tensorflow.keras import Sequential ##GPU
from tensorflow.keras.layers import Dense, LSTM

import math
# from tensorflow import keras ##GPU
# from tensorflow.keras import layers ##GPU
# from tensorflow.keras.layers import Dense, LSTM

from tensorflow.keras import Sequential
from tensorflow.keras.layers import Dense, Activation, Conv1D ,MaxPooling1D, Dropout,LSTM
# from tensorflow.keras.layers import LSTM
from tensorflow.keras.utils import plot_model
from tensorflow.keras.metrics import RootMeanSquaredError as rmse
from tensorflow.keras import optimizers

# define model for GPU
model = Sequential()
model.add(Conv1D(filters=256, kernel_size=2, activation='relu',padding = 'same',input_shape=(120,1)))
model.add(MaxPooling1D(pool_size=2))
model.add(LSTM(100, return_sequences = True)) ##bermasalah
model.add(Dropout(0.3))
model.add(LSTM(100, return_sequences = False)) ##bermasalah
model.add(Dropout(0.3))
model.add(Dense(units=1, activation = 'sigmoid'))  ##aselinya
# model.add(tf.keras.layers.Dense(8, input_shape=(16,))) ##versipanjang

model.compile(optimizer='adam', loss= 'mse' , metrics = [rmse()])

and as far as i know., this line is make me stucked: model.add(LSTM(100, return_sequences = True))

and model.add(LSTM(100, return_sequences = False))

i've also already try to read lstm on tensorflow.. somebody please help me.. :)

Answer 1

2021-12-04 18:02:41.467262: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll

The above mentioned codes starts from I are just (I)information messages and are not errors.

you can surpass (I)informational messages and (W)warnings by using the code below:

import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'

I also replicated the above mentioned code in both CPU and GPU runtime mode in Google colab using TF 2.7, 2.8, which does not show any error.