This zip file contains the 12 files. The first one is this readme file. The files train_0 to train_9 contain the training set for each one of the handwritten digits. The files are in a format that can be used over the libSVM code directly. You will need to link the files together and change the labels from 0, 1, 2, ... to +1 and 1, depending on which type of multi-class SVM you will be training (one-vs-all or one-vs-one). Each file contains approximately 1000 training samples.

The last file you will find is test with the same format as the training set, in this case all the labels are together and it contains 10000 samples.

In case you do not want to use LibSVM, need more information about the pre-processing and want to use the original data, it can be downloaded from my webpage in the software link (prepro.zip) This files contains the original data and C code for preprocessing in. The training set is the first 10000 samples in train and the test set is all the samples in test.
