nemo_rl.data.datasets.response_datasets.megatron_sft_packed#

Module Contents#

Classes#

MegatronSFTPackedDataset

Load Megatron-LM offline-packed SFT JSONL records.

API#

class nemo_rl.data.datasets.response_datasets.megatron_sft_packed.MegatronSFTPackedDataset(
data_path: str,
chat_key: str,
subset: Optional[str] = None,
split: Optional[str] = None,
split_validation_size: float = 0,
seed: int = 42,
context_parallel_size: int | None = None,
**kwargs: Any,
)#

Bases: nemo_rl.data.datasets.raw_dataset.RawDataset

Load Megatron-LM offline-packed SFT JSONL records.

Initialization

format_data(data: dict[str, Any]) → dict[str, Any]#
set_processor() → None#