In reinforcement learning from human feedback, it is common to optimize
...
Over the past two years, EleutherAI has established itself as a radicall...
We introduce GPT-NeoX-20B, a 20 billion parameter autoregressive languag...
This datasheet describes the Pile, a 825 GiB dataset of human-authored t...
Large language models have recently been shown to attain reasonable zero...
While conventional wisdom suggests that more aggressively filtering data...
Recent work has demonstrated that increased training dataset diversity
i...
Storytelling plays a central role in human socializing and entertainment...