Dataset

Dataset for RL-based model and Reward model

Dataset for Reward Model

---

source

PairDataset

 PairDataset (dataset:Iterable, tokenizer:Callable, max_length:int=1024)

Pairwise dataset for train reward model.

	Type	Default	Details
dataset	typing.Iterable		A dataset
tokenizer	typing.Callable		The tokenizer of the reward model
max_length	int	1024	Max context length of the reward model

Dataset for PPO Agent

---

source

PromptDataset

 PromptDataset (dataset:Iterable, tokenizer:Callable, max_length:int=1024)

Dataset for train RL-based language model.

	Type	Default	Details
dataset	typing.Iterable		A dataset
tokenizer	typing.Callable		The tokenizer of the language model
max_length	int	1024	Max context length of the language model