! pip install datasets transformers[sentencepiece]

from datasets import load_dataset

raw_datasets = load_dataset("allocine")
raw_datasets.cache_files

raw_datasets.save_to_disk("my-arrow-datasets")

from datasets import load_from_disk

arrow_datasets_reloaded = load_from_disk("my-arrow-datasets")
arrow_datasets_reloaded

for split, dataset in raw_datasets.items():
    dataset.to_csv(f"my-dataset-{split}.csv", index=None)

data_files = {
    "train": "my-dataset-train.csv",
    "validation": "my-dataset-validation.csv",
    "test": "my-dataset-test.csv",
}

csv_datasets_reloaded = load_dataset("csv", data_files=data_files)
csv_datasets_reloaded

# Save in JSON Lines format
for split, dataset in raw_datasets.items():
    dataset.to_json(f"my-dataset-{split}.jsonl")

# Save in Parquet format
for split, dataset in raw_datasets.items():
    dataset.to_parquet(f"my-dataset-{split}.parquet")

json_data_files = {
    "train": "my-dataset-train.jsonl",
    "validation": "my-dataset-validation.jsonl",
    "test": "my-dataset-test.jsonl",
}

parquet_data_files = {
    "train": "my-dataset-train.parquet",
    "validation": "my-dataset-validation.parquet",
    "test": "my-dataset-test.parquet",
}

# Reload with the `json` script
json_datasets_reloaded = load_dataset("json", data_files=json_data_files)
# Reload with the `parquet` script
parquet_datasets_reloaded = load_dataset("parquet", data_files=parquet_data_files)