sparsity is an R package with functions for sparse matrices.
read.svmlight() and write.svmlight() read/write sparse matrices in SVMlight format.
You will find other functions for this on the internet, but the ones I found were either slow or handled only dense (=normal) matrices.
The LiblineaR CRAN package provides an R interface to the LIBLINEAR library, but uses a dense representation. sparsity's functions use sparse matrices (from the Matrix package) instead. In addition it gives you a pointer to LIBLINEAR's internal representation of the data, which means you can train multiple models without the overhead of transforming the input data.
# install.packages("devtools")
library(devtools)
install_github("sparsity", "felixr")