datadreamer.dev

DataDreamer

Aligning a LLM with Human Preferences# In order to better align the responses instruction-tuned LLMs generate to what humans would prefer, we can train LLMs against a reward model or a dataset of hum