diff --git a/src/plaid/storage/hf_datasets/writer.py b/src/plaid/storage/hf_datasets/writer.py
index 77276c2b..df92818b 100644
--- a/src/plaid/storage/hf_datasets/writer.py
+++ b/src/plaid/storage/hf_datasets/writer.py
@@ -14,10 +14,12 @@
 import gc
 import logging
 import tempfile
+from contextlib import contextmanager
 from pathlib import Path
 from typing import Any, Callable, Generator, Optional, Union
 
 import yaml
+from datasets.utils import logging as datasets_logging
 from huggingface_hub import DatasetCard, hf_hub_download
 
 from plaid.storage.hf_datasets.bridge import generator_to_datasetdict
@@ -29,6 +31,23 @@
 logger = logging.getLogger(__name__)
 
 
+@contextmanager
+def _hf_progress_bars(enabled: bool):
+    """Temporarily configure Hugging Face progress bars."""
+    was_enabled = datasets_logging.is_progress_bar_enabled()
+    if enabled:
+        datasets_logging.enable_progress_bar()
+    else:
+        datasets_logging.disable_progress_bar()
+    try:
+        yield
+    finally:
+        if was_enabled:
+            datasets_logging.enable_progress_bar()
+        else:
+            datasets_logging.disable_progress_bar()
+
+
 def _compute_num_shards(hf_dataset_dict: Any) -> dict[str, int]:
     """Computes the number of shards for each split in a DatasetDict.
 
@@ -76,15 +95,18 @@ def save_datasetdict_to_disk(
         None
     """
     num_shards = _compute_num_shards(hf_datasetdict)
-    num_proc = kwargs.get("num_proc", None)
-    if num_proc is not None:  # pragma: no cover
+    requested_num_proc = kwargs.pop("num_proc", None)
+    if requested_num_proc is None or requested_num_proc <= 1:
+        num_proc = None
+    else:
         min_num_shards = min(num_shards.values())
-        if min_num_shards < num_proc:
+        effective_num_proc = min(requested_num_proc, min_num_shards)
+        num_proc = effective_num_proc if effective_num_proc > 1 else None
+        if effective_num_proc < requested_num_proc:
             logger.warning(
-                f"num_proc changed from {num_proc} to 1 to safely adapt for num_shards={num_shards}"
+                f"num_proc changed from {requested_num_proc} to "
+                f"{effective_num_proc} to safely adapt for num_shards={num_shards}"
             )
-            num_proc = 1
-        del kwargs["num_proc"]
 
     hf_datasetdict.save_to_disk(
         str(Path(path) / "data"), num_shards=num_shards, num_proc=num_proc, **kwargs
@@ -97,7 +119,7 @@ def generate_datasetdict_to_disk(
     variable_schema: dict[str, dict],
     gen_kwargs: Optional[dict[str, dict[str, Any]]] = None,
     num_proc: int = 1,
-    verbose: bool = False,  # noqa: ARG001
+    verbose: bool = False,
 ) -> None:
     """Generates and saves a DatasetDict to disk from sample generators.
 
@@ -111,7 +133,7 @@ def generate_datasetdict_to_disk(
         num_proc (int): Number of processes for generation.
         verbose (bool): Whether to enable verbose output.
     """
-    with tempfile.TemporaryDirectory() as tmpdirname:
+    with _hf_progress_bars(verbose), tempfile.TemporaryDirectory() as tmpdirname:
         hf_datasetdict = generator_to_datasetdict(
             generators,
             variable_schema,
diff --git a/tests/storage/test_hf_datasets_writer.py b/tests/storage/test_hf_datasets_writer.py
new file mode 100644
index 00000000..1cdcd99d
--- /dev/null
+++ b/tests/storage/test_hf_datasets_writer.py
@@ -0,0 +1,65 @@
+from unittest.mock import MagicMock
+
+import pytest
+from datasets.utils import logging as datasets_logging
+
+from plaid.storage.hf_datasets import writer
+
+
+@pytest.mark.parametrize(
+    ("requested_num_proc", "dataset_size", "expected_num_proc"),
+    [
+        (None, 1, None),
+        (1, 1, None),
+        (4, 1, None),
+        (4, 2 * 500 * 1024 * 1024, 2),
+        (2, 4 * 500 * 1024 * 1024, 2),
+    ],
+)
+def test_save_datasetdict_adapts_parallelism(
+    tmp_path, requested_num_proc, dataset_size, expected_num_proc
+):
+    datasetdict = MagicMock()
+    dataset = MagicMock()
+    dataset.__len__.return_value = 10
+    dataset.data.nbytes = dataset_size
+    datasetdict.items.return_value = [("train", dataset)]
+
+    writer.save_datasetdict_to_disk(tmp_path, datasetdict, num_proc=requested_num_proc)
+
+    datasetdict.save_to_disk.assert_called_once_with(
+        str(tmp_path / "data"),
+        num_shards={"train": min(10, max(1, dataset_size // (500 * 1024 * 1024)))},
+        num_proc=expected_num_proc,
+    )
+
+
+@pytest.mark.parametrize("initially_enabled", [False, True])
+@pytest.mark.parametrize("verbose", [False, True])
+def test_generate_controls_and_restores_hf_progress(
+    monkeypatch, tmp_path, initially_enabled, verbose
+):
+    datasetdict = MagicMock()
+    observed = []
+
+    def fake_generate(*_args, **_kwargs):
+        observed.append(datasets_logging.is_progress_bar_enabled())
+        return datasetdict
+
+    monkeypatch.setattr(writer, "generator_to_datasetdict", fake_generate)
+    monkeypatch.setattr(writer, "save_datasetdict_to_disk", MagicMock())
+
+    if initially_enabled:
+        datasets_logging.enable_progress_bar()
+    else:
+        datasets_logging.disable_progress_bar()
+
+    writer.generate_datasetdict_to_disk(
+        tmp_path,
+        generators={"train": MagicMock()},
+        variable_schema={},
+        verbose=verbose,
+    )
+
+    assert observed == [verbose]
+    assert datasets_logging.is_progress_bar_enabled() is initially_enabled