DCLM and self-play->DCLM warm-start checkpoints with training logs

Model card

initial commit