nemo_rl.data.datasets.response_datasets.gsm8k#

Module Contents#

Classes#

GSM8KDataset

Simple wrapper around the GSM8K dataset.

Functions#

_extract_hash_answer

API#

nemo_rl.data.datasets.response_datasets.gsm8k._extract_hash_answer(text: str) → str | None#
class nemo_rl.data.datasets.response_datasets.gsm8k.GSM8KDataset(
subset: str | None = 'main',
split: str = 'train',
extract_answer: bool = True,
**kwargs,
)#

Bases: nemo_rl.data.datasets.raw_dataset.RawDataset

Simple wrapper around the GSM8K dataset.

Parameters:

  • subset – Hugging Face config name, default is “main” (None also selects “main”, since the dataset has no implicit default config). openai/gsm8k also ships a "socratic" config, whose answers interleave Socratic sub-questions with the reasoning steps.

  • split – Split name for the dataset, default is “train”

  • extract_answer – Whether to extract the answer from the dataset, default is True

Initialization

format_data(data: dict[str, Any]) → dict[str, Any]#