cerebras.modelzoo.data.multimodal.llava.config.LlavaHDF5MapDataProcessorConfig#
- class cerebras.modelzoo.data.multimodal.llava.config.LlavaHDF5MapDataProcessorConfig(batch_size: int = <object object at 0x7f9345f8db90>, shuffle: bool = True, shuffle_seed: int = 0, num_workers: int = 0, prefetch_factor: int = 10, persistent_workers: bool = False, img_data_dir: str = <object object at 0x7f9345f8db90>, image_data_size: List[int] = <factory>, transforms: List[dict] = <factory>, data_dir: Union[str, List[str]] = <object object at 0x7f9345f8db90>, use_worker_cache: bool = False, max_sequence_length: Optional[int] = None, mixture: Optional[List[dict]] = None, mixed_precision: Optional[bool] = None, drop_last: bool = True, num_samples: Optional[int] = None, sort_files: bool = True, use_vsl: bool = False, pad_last: bool = False, data_subset: Optional[str] = None, dataset_map_fn: Optional[str] = None)[source]#
- img_data_dir: str = <object object>#
- image_data_size: List[int]#
- transforms: List[dict]#
- data_dir: Union[str, List[str]] = <object object>#
The path to the HDF5 files.
- use_worker_cache: bool = False#
whether or not to copy data to storage that is directly attached to each individual worker node. Useful when your network storage is unusually slow, but otherwise discouraged.
- max_sequence_length: Optional[int] = None#
The sequence length of samples produced by the dataloader. When using the corpus data format, the same preprocessed data will work with any max sequence length, so this may be set at runtime. When using the sample format this must be set to None
- mixture: Optional[List[dict]] = None#
- mixed_precision: Optional[bool] = None#
- drop_last: bool = True#
similar to the PyTorch drop_last setting except that samples that when set to True, samples that would have been dropped at the end of one epoch are yielded at the start of the next epoch so that there is no data loss. This is necessary for a data ordering that is independent of the distributed setup being used.
- num_samples: Optional[int] = None#
- sort_files: bool = True#
whether or not the reader should sort the input files. This is included for backwards compatibility and should almost always be set to True
- use_vsl: bool = False#
Flag to enable variable sequence length training. It requires the dataset to have two extra features
- batch_size: int = <object object>#
Batch size to be used
- pad_last: bool = False#
- shuffle: bool = True#
Whether or not to shuffle the dataset
- shuffle_seed: int = 0#
Seed used for deterministic shuffling
- data_subset: Optional[str] = None#
An optional specification to only consider a subset of the full dataset, useful for sequence length scheduling and multi-epoch testing. Expected to be a comma separated list of ranges, e.g. 0.0-0.5 or 0.1-0.3,0.7-1.0. Specifying 0.0-0.5 creates a dataset from the first half of the data on disk and disregards the second half.
- dataset_map_fn: Optional[str] = None#
- num_workers: int = 0#
The number of PyTorch processes used in the dataloader
- prefetch_factor: int = 10#
The number of batches to prefetch in the dataloader
- persistent_workers: bool = False#
Whether or not to keep workers persistent between epochs