More reward models: Train a DPO model and autoencoders for hh-hrlf

[amirabdullah19852020]

For now, do this just for gpt-125m-neo to make this fast.

Make the pipeline configurable while doing this, so we can easily run this for a range of other models quicker.