Keras: Training in Batches

[GV1028]

I am training large amount of data using a CNN. I am not using a default keras generator but reading images batch by batch in a for loop and in each iteration using model.fit(). The official tutorial tells me that when using model.fit() use nb_epoch = 1. What is the reason for that? If i give nb_epochs as 10 or something. Will the model re initialize the weights during every iteration or will it update the weights.
This is the sample code:

for e in range(234):
x,y = getdata(1000,e)
x = x.astype('float32')
x = x/255
model.fit(x,y,nb_epoch = 10)

getdata() returns 1000 images. So my question is for each 1000 images, does the model update weights or not? Or does it reinitialize weights every iteration.

[fchollet]

Successive calls to fit do not reinitialize the weights.