How to free memory in colab?

Question

I tring to iterate through diffrent hyperparameters to build an optimal model. But after 1 iteration(training of 1 model) is compeleted I'm running out of memory when the 2nd iteration starts.ResourceExhaustedError: OOM when allocating tensor with shape[5877,200,200,3] and type double on /job:localhost/replica:0/task:0/device:GPU:0 by allocator GPU_0_bfc [Op:GatherV2]

I tried using ops.reset_default_graph() but it doen't do anything.

import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf
from tensorflow.keras import regularizers
from tensorflow.keras.layers import Dense,Activation,Flatten,Conv2D,MaxPooling2D,Dropout
import os
import cv2
import random
import pickle
import time
from tensorflow.keras.utils import to_categorical
from tensorflow.keras.callbacks import TensorBoard
from google.colab import files
from tensorflow.python.framework import ops
p1=open("/content/tfds.pickle","rb")
def prepare_ds():
    dir="drive//My Drive//dataset//"
    cat=os.listdir(dir)
    i=1
    td=[]
    for x in cat:
        d=dir+x
        y1=cat.index(x)
        for img in os.listdir(d):
            im=cv2.imread(d+"//"+img)
            print(i)
            i=i+1     
            im=cv2.resize(im,(200,200))
            td.append([im,y1])
    ##      im[:,:,0],im[:,:,2]=im[:,:,2],im[:,:,0].copy()
    ##      plt.imshow(im)
    ##      plt.show()
    random.shuffle(td)
    X=[]
    Y=[]
    for a1,a2 in td:
        X.append(a1)
        Y.append(a2)
    X=np.array(X).reshape(-1,200,200,3)
    Y=np.array(Y).reshape(-1,1)
    pickle.dump([X,Y],p1)
##prepare_ds()
X,Y=pickle.load(p1)
X=X/255.0
def learn():
    model=tf.keras.models.Sequential()
    model.add(Conv2D(lsi,(3,3),input_shape=X.shape[1:]))
    model.add(Activation("relu"))
    model.add(MaxPooling2D(pool_size=(2,2)))

    for l in range(cli-1):
      model.add(Conv2D(lsi,(3,3)))
      model.add(Activation("relu"))
      model.add(MaxPooling2D(pool_size=(2,2)))

    model.add(Flatten())
    for l in range(dli):
      model.add(Dense(lsi))
      model.add(Activation("relu"))

    model.add(Dropout(0.5))
    model.add(Dense(10))
    model.add(Activation('softmax'))

    model.compile(loss="sparse_categorical_crossentropy",optimizer="adam",metrics=['accuracy'])
    model.fit(X,Y,batch_size=16,validation_split=0.1,epochs=3,verbose=2,callbacks=[tb])
    model.save('tm1.h5')
    ops.reset_default_graph()

dl=[0,1,2]
ls=[32,64,128]
cl=[1,2,3]
for dli in dl:
  for lsi in ls:
    for cli in cl:
      ops.reset_default_graph()
      NAME = "{}-conv-{}-nodes-{}-dense".format(cli, lsi, dli)
      tb=TensorBoard(log_dir="logs//{}".format(NAME))
      print(NAME)
      learn()

p1.close()
!zip -r /content/file.zip /content/logs
!cp file.zip "/content/drive/My Drive/"

Answer 1

Hi there.

You can use the built-in Garbage Collector library in Python. I often create a custom callback that uses this library on the end of each epoch. You can think of it as clearing cached information you no longer need

# Garbage Collector - use it like gc.collect()
import gc

# Custom Callback To Include in Callbacks List At Training Time
class GarbageCollectorCallback(tf.keras.callbacks.Callback):
    def on_epoch_end(self, epoch, logs=None):
        gc.collect()

Additionally just try running the command gc.collect() by itself to see the results and see how it works. Here is some documentation on how it works. I often use it to keep my kernel sizes small in kernel only Kaggle competitions**

I hope this helps!