feat(dataset): integrate episode streaming into training

This commit is contained in:
Pepijn
2026-07-23 16:10:39 +02:00
parent b85620657f
commit 3d70b21aac
34 changed files with 2771 additions and 942 deletions
+13
View File
@@ -36,3 +36,16 @@ def test_dataset_config_none_episodes_ok():
def test_dataset_config_empty_episodes_ok():
DatasetConfig(repo_id="user/repo", episodes=[])
@pytest.mark.parametrize(
("field", "value", "message"),
[
("streaming_episode_pool_size", 0, "episode_pool_size"),
("streaming_prefetch_episodes", -1, "prefetch_episodes"),
("streaming_byte_budget_gb", 0, "byte_budget_gb"),
],
)
def test_dataset_config_rejects_invalid_streaming_resource_limits(field, value, message):
with pytest.raises(ValueError, match=message):
DatasetConfig(repo_id="user/repo", **{field: value})