Struct HuggingfaceDatasetLoader 

pub struct HuggingfaceDatasetLoader { /* private fields */ }

Available on crate features sqlite or sqlite-bundled only.

Load a dataset from huggingface datasets.

The dataset with all splits is stored in a single sqlite database (see SqliteDataset).

Example

 use burn_dataset::HuggingfaceDatasetLoader;
 use burn_dataset::SqliteDataset;
 use serde::{Deserialize, Serialize};

#[derive(Deserialize, Debug, Clone)]
struct MnistItemRaw {
    pub image_bytes: Vec<u8>,
    pub label: usize,
}

 let train_ds:SqliteDataset<MnistItemRaw> = HuggingfaceDatasetLoader::new("mnist")
      .dataset("train")
      .unwrap();

Note

This loader relies on the datasets library by HuggingFace to download datasets. This is a Python library, so you must have an existing Python installation.

Implementations

impl HuggingfaceDatasetLoader

pub fn new(name: &str) -> Self

Create a huggingface dataset loader.

pub fn with_subset(self, subset: &str) -> Self

Create a huggingface dataset loader for a subset of the dataset.

The subset name must be one of the subsets listed in the dataset page.

If no subset names are listed, then do not use this method.

pub fn with_base_dir(self, base_dir: &str) -> Self

Specify a base directory to store the dataset.

If not specified, the dataset will be stored in ~/.cache/burn-dataset.

pub fn with_huggingface_token(self, huggingface_token: &str) -> Self

Specify a huggingface token to download datasets behind authentication.

You can get a token from tokens settings

pub fn with_huggingface_cache_dir(self, huggingface_cache_dir: &str) -> Self

Specify a huggingface cache directory to store the downloaded datasets.

If not specified, the dataset will be stored in ~/.cache/huggingface/datasets.

pub fn with_huggingface_data_dir(self, huggingface_data_dir: &str) -> Self

Specify a relative path to a subset of a dataset. This is used in some datasets for the manual steps of dataset download process.

Unless you’ve encountered a ManualDownloadError when loading your dataset you probably don’t have to worry about this setting.

pub fn with_trust_remote_code(self, trust_remote_code: bool) -> Self

Specify whether or not to trust remote code.

If not specified, trust remote code is set to true.

pub fn with_use_python_venv(self, use_python_venv: bool) -> Self

Specify whether or not to use the burn-dataset Python virtualenv for running the importer script. If false, local python3’s environment is used.

If not specified, the virtualenv is used.

pub fn dataset<I: DeserializeOwned + Clone>( self, split: &str, ) -> Result<SqliteDataset<I>, ImporterError>

Load the dataset.

pub fn db_file(self) -> Result<PathBuf, ImporterError>

Get the path to the sqlite database file.

If the database file does not exist, it will be downloaded and imported.