From 392f2ba87b2c8e33db159a53c6fba2c3e921f5ee Mon Sep 17 00:00:00 2001
From: Hashim Sharif <hsharif3@tyler.cs.illinois.edu>
Date: Sun, 26 Jul 2020 23:40:25 -0500
Subject: [PATCH] Keras: Creating top level class for all Benchmarks
---
llvm/projects/keras/src/Benchmark.py | 25 +++
llvm/projects/keras/src/__init__.py | 0
llvm/projects/keras/src/alexnet.py | 280 ++++++++++++++-------------
3 files changed, 169 insertions(+), 136 deletions(-)
create mode 100644 llvm/projects/keras/src/Benchmark.py
create mode 100644 llvm/projects/keras/src/__init__.py
diff --git a/llvm/projects/keras/src/Benchmark.py b/llvm/projects/keras/src/Benchmark.py
new file mode 100644
index 0000000000..f37e4de6bc
--- /dev/null
+++ b/llvm/projects/keras/src/Benchmark.py
@@ -0,0 +1,25 @@
+
+
+
+class Benchmark:
+
+ def __init__(self):
+ return
+
+ def buildModel(self):
+ return
+
+ def data_preprocess(self):
+ return
+
+ def trainModel(self):
+ return
+
+ def inference(self):
+ return
+
+ def run(self):
+ return
+
+
+
diff --git a/llvm/projects/keras/src/__init__.py b/llvm/projects/keras/src/__init__.py
new file mode 100644
index 0000000000..e69de29bb2
diff --git a/llvm/projects/keras/src/alexnet.py b/llvm/projects/keras/src/alexnet.py
index 28bca02e89..d10bfa6b9b 100644
--- a/llvm/projects/keras/src/alexnet.py
+++ b/llvm/projects/keras/src/alexnet.py
@@ -18,176 +18,182 @@ import struct
import keras
import numpy as np
import os
+from Benchmark import Benchmark
from frontend.approxhpvm_translator import translate_to_approxhpvm
from frontend.weight_utils import dumpCalibrationData
from frontend.weight_utils import dumpHPVMToKerasModel
-def lr_schedule(epoch):
- lrate = 0.001
- if epoch > 20:
- lrate = 0.0005
- if epoch > 40:
- lrate = 0.0003
- if epoch > 60:
- lrate = 0.0001
- if epoch > 80:
- lrate = 0.00005
-
- return lrate
+class AlexNet(Benchmark):
-def buildModel():
+ def __init__(self):
+ self.name = "AlexNet"
+
+
+ def lr_schedule(self, epoch):
+ lrate = 0.001
+ if epoch > 20:
+ lrate = 0.0005
+ if epoch > 40:
+ lrate = 0.0003
+ if epoch > 60:
+ lrate = 0.0001
+ if epoch > 80:
+ lrate = 0.00005
- activation_type = "tanh"
- weight_decay = 1e-4
-
- model = Sequential()
- model.add(Conv2D(64, kernel_size=(11, 11), activation=activation_type,
- input_shape=(3, 32, 32), padding = 'same',
- kernel_regularizer=regularizers.l2(weight_decay) ))
- model.add(MaxPooling2D(pool_size=(2, 2), strides=(2,2) ))
- model.add(Dropout(0.2))
- model.add(Conv2D(192, kernel_size=(5, 5), activation=activation_type, padding = 'same',
- kernel_regularizer=regularizers.l2(weight_decay)))
- model.add(MaxPooling2D(pool_size=(2, 2), strides=(2,2) ))
- model.add(Dropout(0.3))
-
- model.add(Conv2D(384, kernel_size=(3, 3), activation=activation_type, padding = 'same',
- kernel_regularizer=regularizers.l2(weight_decay) ))
- model.add(Conv2D(256, kernel_size=(3, 3), activation=activation_type, padding = 'same',
- kernel_regularizer=regularizers.l2(weight_decay) ))
- model.add(Conv2D(256, kernel_size=(3, 3), activation=activation_type, padding = 'same',
- kernel_regularizer=regularizers.l2(weight_decay) ))
- model.add(MaxPooling2D(pool_size=(2, 2), strides=(2,2) ))
- model.add(Dropout(0.4))
-
- model.add(Flatten())
- #model.add(Flatten())
- #model.add(Dense(256))
- model.add(Dense(10))
- model.add(Activation('softmax'))
-
- return model
+ return lrate
-def buildModel_old():
-
- model = Sequential()
- model.add(Conv2D(128, kernel_size=(3, 3), activation='tanh', input_shape=(3, 32, 32), padding = 'same'))
- model.add(Conv2D(256, kernel_size=(3, 3), activation='tanh', padding = 'same'))
- model.add(MaxPooling2D(pool_size=(2, 2)))
- #model.add(Dropout(0.25))
-
- model.add(Conv2D(256, kernel_size=(3, 3), activation='tanh', padding = 'same'))
- model.add(MaxPooling2D(pool_size=(2, 2)))
-
- model.add(Conv2D(256, kernel_size=(3, 3), activation='tanh', padding = 'same'))
- model.add(MaxPooling2D(pool_size=(2, 2)))
- model.add(Conv2D(256, kernel_size=(3, 3), activation='tanh', padding = 'same'))
- model.add(MaxPooling2D(pool_size=(2, 2)))
- #model.add(Dropout(0.25))
-
- model.add(Flatten())
- #model.add(Flatten())
- model.add(Dense(4096, activation='tanh'))
- #model.add(Dropout(0.5))
- model.add(Dense(2048, activation='tanh'))
- model.add(Dense(10, activation='tanh'))
- model.add(Activation('softmax'))
-
- return model
+ def buildModel(self):
+ activation_type = "tanh"
+ weight_decay = 1e-4
+ model = Sequential()
+ model.add(Conv2D(64, kernel_size=(11, 11), activation=activation_type,
+ input_shape=(3, 32, 32), padding = 'same',
+ kernel_regularizer=regularizers.l2(weight_decay) ))
+ model.add(MaxPooling2D(pool_size=(2, 2), strides=(2,2) ))
+ model.add(Dropout(0.2))
+ model.add(Conv2D(192, kernel_size=(5, 5), activation=activation_type, padding = 'same',
+ kernel_regularizer=regularizers.l2(weight_decay)))
+ model.add(MaxPooling2D(pool_size=(2, 2), strides=(2,2) ))
+ model.add(Dropout(0.3))
+ model.add(Conv2D(384, kernel_size=(3, 3), activation=activation_type, padding = 'same',
+ kernel_regularizer=regularizers.l2(weight_decay) ))
+ model.add(Conv2D(256, kernel_size=(3, 3), activation=activation_type, padding = 'same',
+ kernel_regularizer=regularizers.l2(weight_decay) ))
+ model.add(Conv2D(256, kernel_size=(3, 3), activation=activation_type, padding = 'same',
+ kernel_regularizer=regularizers.l2(weight_decay) ))
+ model.add(MaxPooling2D(pool_size=(2, 2), strides=(2,2) ))
+ model.add(Dropout(0.4))
+
+ model.add(Flatten())
+ #model.add(Flatten())
+ #model.add(Dense(256))
+ model.add(Dense(10))
+ model.add(Activation('softmax'))
+
+ return model
+
+
+
+ def buildModel_old():
+
+ model = Sequential()
+ model.add(Conv2D(128, kernel_size=(3, 3), activation='tanh', input_shape=(3, 32, 32), padding = 'same'))
+ model.add(Conv2D(256, kernel_size=(3, 3), activation='tanh', padding = 'same'))
+ model.add(MaxPooling2D(pool_size=(2, 2)))
+ #model.add(Dropout(0.25))
+
+ model.add(Conv2D(256, kernel_size=(3, 3), activation='tanh', padding = 'same'))
+ model.add(MaxPooling2D(pool_size=(2, 2)))
+
+ model.add(Conv2D(256, kernel_size=(3, 3), activation='tanh', padding = 'same'))
+ model.add(MaxPooling2D(pool_size=(2, 2)))
+ model.add(Conv2D(256, kernel_size=(3, 3), activation='tanh', padding = 'same'))
+ model.add(MaxPooling2D(pool_size=(2, 2)))
+ #model.add(Dropout(0.25))
+
+ model.add(Flatten())
+ #model.add(Flatten())
+ model.add(Dense(4096, activation='tanh'))
+ #model.add(Dropout(0.5))
+ model.add(Dense(2048, activation='tanh'))
+ model.add(Dense(10, activation='tanh'))
+ model.add(Activation('softmax'))
+
+ return model
-def trainModel(model):
- (X_train, Y_train), (X_test, Y_test) = cifar10.load_data()
- test_labels = Y_test
- train_labels = Y_train
- #X_train = X_train.astype('float32')
- #X_test = X_test.astype('float32')
- X_train = X_train / 255.0
- X_test = X_test / 255.0
-
- mean = np.mean(X_train,axis=(0,1,2,3))
- std = np.std(X_train,axis=(0,1,2,3))
- X_train = (X_train-mean)/(std+1e-7)
- X_test = (X_test-mean)/(std+1e-7)
-
- dir_prefix = "/home/hsharif3/Gitlab/hpvm/llvm/projects/hpvm-tensor-rt/model_params/alexnet_cifar10/"
- #opt_rms = keras.optimizers.rmsprop(lr=0.001,decay=1e-6)
- # Compile the model
- model.compile(loss='categorical_crossentropy',
- optimizer=Adam(lr=0.0001, decay=1e-6),
- #optimizer = opt_rms,
- metrics=['accuracy'])
+ def trainModel(self, model):
- #print to_categorical(Y_train, 10)
- print (to_categorical(Y_train))
+ (X_train, Y_train), (X_test, Y_test) = cifar10.load_data()
+ test_labels = Y_test
+ train_labels = Y_train
+ #X_train = X_train.astype('float32')
+ #X_test = X_test.astype('float32')
+ X_train = X_train / 255.0
+ X_test = X_test / 255.0
- datagen = ImageDataGenerator(
+ mean = np.mean(X_train,axis=(0,1,2,3))
+ std = np.std(X_train,axis=(0,1,2,3))
+ X_train = (X_train-mean)/(std+1e-7)
+ X_test = (X_test-mean)/(std+1e-7)
+
+ dir_prefix = "/home/hsharif3/Gitlab/hpvm/llvm/projects/hpvm-tensor-rt/model_params/alexnet_cifar10/"
+
+ #opt_rms = keras.optimizers.rmsprop(lr=0.001,decay=1e-6)
+ # Compile the model
+ model.compile(loss='categorical_crossentropy',
+ optimizer=Adam(lr=0.0001, decay=1e-6),
+ #optimizer = opt_rms,
+ metrics=['accuracy'])
+
+ #print to_categorical(Y_train, 10)
+ print (to_categorical(Y_train))
+
+
+ datagen = ImageDataGenerator(
rotation_range=15,
width_shift_range=0.1,
height_shift_range=0.1,
horizontal_flip=True,
- )
- datagen.fit(X_train)
+ )
+ datagen.fit(X_train)
-
- model.fit(X_train, to_categorical(Y_train, 10),
- batch_size=128,
- shuffle=True,
- epochs = 1,
- #epochs=100,
- validation_data=(X_test, to_categorical(Y_test, 10)), callbacks=[LearningRateScheduler(lr_schedule)])
-
- # Evaluate the model
- scores = model.evaluate(X_test, to_categorical(Y_test, 10))
-
- print('Loss: %.3f' % scores[0])
- print('Accuracy: %.3f' % scores[1])
- print ("*** TRAINED MODEL ****\n")
-
- #dumpCalibrationData("calibration_data/alexnet_calib.bin", X_train,
- # "calibration_data/alexnet_train_labels.bin", train_labels)
+ model.fit(X_train, to_categorical(Y_train, 10),
+ batch_size=128,
+ shuffle=True,
+ epochs = 1,
+ #epochs=100,
+ validation_data=(X_test, to_categorical(Y_test, 10)), callbacks=[LearningRateScheduler(self.lr_schedule)])
+ # Evaluate the model
+ scores = model.evaluate(X_test, to_categorical(Y_test, 10))
-
+ print('Loss: %.3f' % scores[0])
+ print('Accuracy: %.3f' % scores[1])
+ print ("*** TRAINED MODEL ****\n")
-def reloadKerasModel(model_path):
+ return model
- model = load_model(model_path)
-
- score = model.evaluate(X_test, to_categorical(Y_test, 10), verbose=0)
- print('Test loss2:', score[0])
- print('Test accuracy2:', score[1])
-
-def data_preprocess():
+ def reloadKerasModel(model_path):
- (X_train, Y_train), (X_test, Y_test) = cifar10.load_data()
+ model = load_model(model_path)
+
+ score = model.evaluate(X_test, to_categorical(Y_test, 10), verbose=0)
+ print('Test loss2:', score[0])
+ print('Test accuracy2:', score[1])
+
+
+
+ def data_preprocess(self):
+
+ (X_train, Y_train), (X_test, Y_test) = cifar10.load_data()
+
+ X_train = X_train / 255.0
+ X_test = X_test / 255.0
+
+ mean = np.mean(X_train,axis=(0,1,2,3))
+ std = np.std(X_train,axis=(0,1,2,3))
+ X_train = (X_train-mean)/(std+1e-7)
+ X_test = (X_test-mean)/(std+1e-7)
+
+ return X_train, Y_train, X_test, Y_test
- X_train = X_train / 255.0
- X_test = X_test / 255.0
-
- mean = np.mean(X_train,axis=(0,1,2,3))
- std = np.std(X_train,axis=(0,1,2,3))
- X_train = (X_train-mean)/(std+1e-7)
- X_test = (X_test-mean)/(std+1e-7)
- return X_train, Y_train, X_test, Y_test
-
-
if __name__ == "__main__":
@@ -199,9 +205,11 @@ if __name__ == "__main__":
# Changing to NCHW format
K.set_image_data_format('channels_first')
- model = buildModel()
+ alexnet = AlexNet()
+
+ model = alexnet.buildModel()
- X_train, Y_train, X_test, Y_test = data_preprocess()
+ X_train, Y_train, X_test, Y_test = alexnet.data_preprocess()
reload_dir = "/home/hsharif3/Gitlab/hpvm/llvm/projects/hpvm-tensor-rt/model_params/alexnet_cifar10/"
keras_model_file = "alexnet.h5"
@@ -212,7 +220,7 @@ if __name__ == "__main__":
model = load_model(keras_model_file)
if sys.argv[1] == "train":
- model = trainModel(model)
+ model = alexnet.trainModel(model)
num_classes = 10
score = model.evaluate(X_test, to_categorical(Y_test, num_classes), verbose=0)
@@ -220,7 +228,7 @@ if __name__ == "__main__":
if len(sys.argv) > 2 and sys.argv[2] == "frontend":
- if sys.argv[1] != "hpvm_reload":
+ if sys.argv[1] == "keras_reload":
print("ERROR: Must load HPVM model to invoke frontend")
sys.exit(1)
--
GitLab