File size: 8,869 Bytes
0185029 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 | from __future__ import annotations
import hashlib
import json
import sys
import tempfile
import unittest
from pathlib import Path
import yaml
ORARL_ROOT = Path(__file__).resolve().parents[1]
if str(ORARL_ROOT) not in sys.path:
sys.path.insert(0, str(ORARL_ROOT))
from orarl.data.build import ConfigError, ShortfallError, build_dataset # noqa: E402
def _write_jsonl(path: Path, rows: list[dict[str, object]]) -> None:
with path.open("w", encoding="utf-8") as handle:
for row in rows:
handle.write(json.dumps(row, sort_keys=True) + "\n")
def _fixture_config(tmp_path: Path, quota: int = 2, canary_size: int = 1) -> Path:
media = tmp_path / "media"
media.mkdir()
for name in ("a.mp4", "b.mp4", "c.mp4", "benchmark.mp4"):
(media / name).write_bytes(name.encode("utf-8"))
_write_jsonl(
tmp_path / "source.jsonl",
[
{"problem": "Prompt one", "answer": "A", "videos": ["a.mp4"]},
{"problem": " PROMPT ONE ", "answer": "A2", "videos": ["./a.mp4"]},
{"problem": "Prompt two", "answer": "B", "videos": ["b.mp4"]},
{"problem": "Prompt three", "answer": "C", "videos": ["c.mp4"]},
{
"problem": "Training prompt on benchmark media",
"answer": "D",
"videos": ["benchmark.mp4"],
},
],
)
_write_jsonl(
tmp_path / "benchmarks.jsonl",
[
{
"problem": "Different benchmark prompt",
"path": str(media / "benchmark.mp4"),
"problem_type": "video_qa_mc",
}
],
)
config = {
"seed": 17,
"target": quota,
"canary_size": canary_size,
"max_prompts_per_media": 1,
"require_media": True,
"benchmark_excludes": ["benchmarks.jsonl"],
"sources": [
{
"name": "video_source",
"input": "source.jsonl",
"task": "video_qa_mc",
"family": "video_qa",
"quota": quota,
"media_root": "media",
"license": "test-only",
}
],
}
config_path = tmp_path / f"config-{quota}-{canary_size}.json"
config_path.write_text(json.dumps(config), encoding="utf-8")
return config_path
class DataBuildTests(unittest.TestCase):
def test_example_manifest_matches_materialized_paper_mixture(self) -> None:
payload = yaml.safe_load(
(ORARL_ROOT / "configs" / "data_sources.example.yaml").read_text(encoding="utf-8")
)
quotas = {source["name"]: source["quota"] for source in payload["sources"]}
self.assertEqual(payload["target"], 100_032)
self.assertEqual(
quotas,
{
"temporal_grounding_train": 20_096,
"tracking_train": 13_952,
"segmentation_train": 12_032,
"spatial_grounding_train": 7_040,
"spatial_temporal_grounding_train": 9_536,
"video_qa_train": 20_288,
"spatial_intelligence_train": 17_088,
},
)
self.assertEqual(sum(quotas.values()), payload["target"])
def test_build_is_deterministic_and_audited(self) -> None:
with tempfile.TemporaryDirectory() as raw_tmp:
tmp_path = Path(raw_tmp)
config = _fixture_config(tmp_path)
first_train = tmp_path / "first.jsonl"
first_canary = tmp_path / "first.canary.jsonl"
first_manifest_path = tmp_path / "first.manifest.json"
second_train = tmp_path / "second.jsonl"
second_canary = tmp_path / "second.canary.jsonl"
manifest = build_dataset(
config,
first_train,
canary_output=first_canary,
manifest_output=first_manifest_path,
)
build_dataset(
config,
second_train,
canary_output=second_canary,
manifest_output=tmp_path / "second.manifest.json",
)
self.assertEqual(first_train.read_bytes(), second_train.read_bytes())
self.assertEqual(first_canary.read_bytes(), second_canary.read_bytes())
self.assertEqual(manifest["actual_train_rows"], 2)
self.assertEqual(manifest["actual_canary_rows"], 1)
self.assertEqual(manifest["counts"]["train"]["by_task"], {"video_qa_mc": 2})
self.assertEqual(manifest["counts"]["train"]["by_source"], {"video_source": 2})
self.assertEqual(manifest["duplicates"]["candidate_prompt_rows"], 1)
self.assertEqual(manifest["leakage"]["candidate_rows_excluded"], 1)
self.assertTrue(all(value == 0 for value in manifest["leakage"]["output"].values()))
self.assertEqual(
manifest["checksums"]["train"],
hashlib.sha256(first_train.read_bytes()).hexdigest(),
)
self.assertEqual(
manifest["checksums"]["canary"],
hashlib.sha256(first_canary.read_bytes()).hexdigest(),
)
def test_short_quota_fails_before_writing(self) -> None:
with tempfile.TemporaryDirectory() as raw_tmp:
tmp_path = Path(raw_tmp)
config = _fixture_config(tmp_path, quota=4, canary_size=0)
train = tmp_path / "short.jsonl"
with self.assertRaisesRegex(ShortfallError, "video_source short by 1"):
build_dataset(config, train)
self.assertFalse(train.exists())
self.assertFalse((tmp_path / "short.canary.jsonl").exists())
self.assertFalse((tmp_path / "short.manifest.json").exists())
def test_short_quota_can_be_explicitly_allowed(self) -> None:
with tempfile.TemporaryDirectory() as raw_tmp:
tmp_path = Path(raw_tmp)
config = _fixture_config(tmp_path, quota=4, canary_size=0)
train = tmp_path / "allowed.jsonl"
manifest = build_dataset(config, train, allow_shortfall=True)
self.assertEqual(manifest["actual_train_rows"], 3)
self.assertEqual(
manifest["selection"]["source_shortfalls"],
{"video_source": 1},
)
def test_preserves_declared_scoring_subtypes_when_requested(self) -> None:
with tempfile.TemporaryDirectory() as raw_tmp:
tmp_path = Path(raw_tmp)
media = tmp_path / "media"
media.mkdir()
for name in ("one.mp4", "two.mp4"):
(media / name).write_bytes(b"media")
_write_jsonl(
tmp_path / "spatial.jsonl",
[
{
"problem": "How many objects?",
"answer": "2",
"videos": ["one.mp4"],
"problem_type": "object_counting",
},
{
"problem": "Which direction?",
"answer": "A",
"videos": ["two.mp4"],
"problem_type": "object_rel_direction",
},
],
)
config = {
"target": 2,
"canary_size": 0,
"require_media": True,
"sources": [
{
"name": "spatial_intelligence",
"input": "spatial.jsonl",
"task": "spatial intelligence",
"family": "spatial_intelligence",
"preserve_problem_type": True,
"quota": 2,
"media_root": "media",
}
],
}
config_path = tmp_path / "config.json"
config_path.write_text(json.dumps(config), encoding="utf-8")
output = tmp_path / "train.jsonl"
build_dataset(config_path, output)
rows = [json.loads(line) for line in output.read_text(encoding="utf-8").splitlines()]
self.assertEqual(
{row["problem_type"] for row in rows},
{"object_counting", "object_rel_direction"},
)
def test_output_cannot_replace_an_input(self) -> None:
with tempfile.TemporaryDirectory() as raw_tmp:
tmp_path = Path(raw_tmp)
config = _fixture_config(tmp_path)
with self.assertRaisesRegex(ConfigError, "cannot replace"):
build_dataset(
config,
tmp_path / "source.jsonl",
overwrite=True,
)
if __name__ == "__main__":
unittest.main()
|