Question about LRM model training

[MuL1ian]

Dear author, I'm a bit confused with the training procedure of LRM, there is a graph indicating after and before RL performance, are we first SFT the QWQ model and then RL, if so will we release the SFT dataset?

[samjia2000]

We directly fine-tune the QwQ model without running SFT. So there is no SFT dataset actually.