I don't have the specific code on me but you should look into the lavaan package (you can specify this sort of estimation of your model) and if you are specifically looking at something like IRT lavaan will do this but you will have to calculate all the parameter estimates by hand (possible but time consuming) there is a package that I know of but haven't used much called ltm (for latent trait modeling)....  both of these packages can be used with marginal maximum likelihood  or WLS-MV.

I hope that is helpful, I can post more details later (if you think it would help) when I have all my resources at hand.

Thank you very much for your answer.

I'm a little bit suprised because I thought lavaan will only has implemented covariance-based estimators and the PMLPLS-PM estimator is a component-based estimator... But I will have a look.

Thanks again!

FYI: The SAS code is provided by Christian Derquenne personally...