Add support for revision dataset parameter

This commit is contained in:
Thomas Cleberg
2024-09-12 16:34:58 -05:00
committed by Wing Lian
parent 2fbc6b0c64
commit 68db5b1b67
3 changed files with 77 additions and 1 deletions

View File

@@ -267,6 +267,73 @@ class TestDatasetPreparation(unittest.TestCase):
assert "attention_mask" in dataset.features
assert "labels" in dataset.features
def test_load_hub_with_revision(self):
"""Verify that processing data from the hub works with a specific revision"""
with tempfile.TemporaryDirectory() as tmp_dir:
prepared_path = Path(tmp_dir) / "prepared"
cfg = DictDefault(
{
"tokenizer_config": "huggyllama/llama-7b",
"sequence_len": 1024,
"datasets": [
{
"path": "mhenrichsen/alpaca_2k_test",
"type": "alpaca",
"revision": "foo",
},
],
}
)
dataset, _ = load_tokenized_prepared_datasets(
self.tokenizer, cfg, prepared_path
)
assert len(dataset) == 2000
assert "input_ids" in dataset.features
assert "attention_mask" in dataset.features
assert "labels" in dataset.features
def test_load_local_hub_with_revision(self):
"""Verify that a local copy of a hub dataset can be loaded with a specific revision"""
with tempfile.TemporaryDirectory() as tmp_dir:
tmp_ds_path = Path("mhenrichsen/alpaca_2k_test")
tmp_ds_path.mkdir(parents=True, exist_ok=True)
snapshot_download(
repo_id="mhenrichsen/alpaca_2k_test",
repo_type="dataset",
local_dir=tmp_ds_path,
revision="foo",
)
prepared_path = Path(tmp_dir) / "prepared"
cfg = DictDefault(
{
"tokenizer_config": "huggyllama/llama-7b",
"sequence_len": 1024,
"datasets": [
{
"path": "mhenrichsen/alpaca_2k_test",
"ds_type": "parquet",
"type": "alpaca",
"data_files": [
"mhenrichsen/alpaca_2k_test/alpaca_2000.parquet",
],
"revision": "foo",
},
],
}
)
dataset, _ = load_tokenized_prepared_datasets(
self.tokenizer, cfg, prepared_path
)
assert len(dataset) == 2000
assert "input_ids" in dataset.features
assert "attention_mask" in dataset.features
assert "labels" in dataset.features
shutil.rmtree(tmp_ds_path)
if __name__ == "__main__":
unittest.main()