Image-Text-to-Text
Transformers
Safetensors
qwen3_5
vllm
video
multimodal
reinforcement-learning
temporal-grounding
object-tracking
video-segmentation
visual-question-answering
spatial-reasoning
qwen3.5
conversational
Instructions to use OraRL/Video-ORA-4B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use OraRL/Video-ORA-4B with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("image-text-to-text", model="OraRL/Video-ORA-4B") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] pipe(text=messages)# Load model directly from transformers import AutoProcessor, AutoModelForMultimodalLM processor = AutoProcessor.from_pretrained("OraRL/Video-ORA-4B") model = AutoModelForMultimodalLM.from_pretrained("OraRL/Video-ORA-4B", device_map="auto") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] inputs = processor.apply_chat_template( messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt", ).to(model.device) outputs = model.generate(**inputs, max_new_tokens=40) print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:])) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use OraRL/Video-ORA-4B with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "OraRL/Video-ORA-4B" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "OraRL/Video-ORA-4B", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Describe this image in one sentence." }, { "type": "image_url", "image_url": { "url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg" } } ] } ] }'Use Docker
docker model run hf.co/OraRL/Video-ORA-4B
- SGLang
How to use OraRL/Video-ORA-4B with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "OraRL/Video-ORA-4B" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "OraRL/Video-ORA-4B", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Describe this image in one sentence." }, { "type": "image_url", "image_url": { "url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg" } } ] } ] }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "OraRL/Video-ORA-4B" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "OraRL/Video-ORA-4B", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Describe this image in one sentence." }, { "type": "image_url", "image_url": { "url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg" } } ] } ] }' - Docker Model Runner
How to use OraRL/Video-ORA-4B with Docker Model Runner:
docker model run hf.co/OraRL/Video-ORA-4B
File size: 15,710 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 | from __future__ import annotations
import json
import sys
import types
from pathlib import Path
import pytest
import yaml
import orarl.evaluation.staging as evaluation_staging
from orarl.cli import eval_data
from orarl.evaluation import (
StagingError,
UploadError,
build_evaluation_repository,
dataset_card_metadata,
dataset_card_subsets,
export_evaluation_index,
export_public_evaluation_repository,
inventory_evaluation_sources,
upload_evaluation_repository,
)
def _write_jsonl(path: Path, records: list[dict[str, object]]) -> None:
path.parent.mkdir(parents=True, exist_ok=True)
path.write_text(
"".join(
json.dumps(record, ensure_ascii=False, sort_keys=True) + "\n" for record in records
),
encoding="utf-8",
)
def _source_fixture(
tmp_path: Path,
*,
expected_count: int | None = 1,
with_artifact: bool = False,
) -> Path:
video_root = tmp_path / "source-videos"
video_root.mkdir()
(video_root / "clip.mp4").write_bytes(b"release-video")
artifact_root = tmp_path / "source-artifacts"
if with_artifact:
artifact_root.mkdir()
(artifact_root / "clip.npz").write_bytes(b"processed-video")
annotation = tmp_path / "annotations.jsonl"
row: dict[str, object] = {
"id": "sample-1",
"question": "What happens?",
"answer": "A",
"video": "clip.mp4",
}
if with_artifact:
row["preprocessed_video"] = "clip.npz"
_write_jsonl(
annotation,
[row],
)
manifest = tmp_path / "private-sources.jsonl"
source: dict[str, object] = {
"benchmark": "videomme",
"eval_task": "videomme",
"split": "test",
"family": "video_qa",
"adapter": "generic",
"annotation_input": str(annotation),
"video_root": str(video_root),
"expected_count": expected_count,
"license": "fixture-only",
"source_url": "https://example.org/videomme",
"redistribution_authorized": True,
"evaluation": {
"prompt_profile": "fixture",
"parser_profile": "multiple_choice",
"metric_profile": "accuracy",
},
}
if with_artifact:
source["preprocessed_root"] = str(artifact_root)
_write_jsonl(
manifest,
[source],
)
return manifest
def _card_header(path: Path) -> dict[str, object]:
text = path.read_text(encoding="utf-8")
assert text.startswith("---\n")
_prefix, header, _body = text.split("---\n", 2)
value = yaml.safe_load(header)
assert isinstance(value, dict)
return value
def test_build_generates_deterministic_hf_card_and_attributes(tmp_path: Path) -> None:
manifest = _source_fixture(tmp_path)
first = tmp_path / "first"
second = tmp_path / "second"
build_evaluation_repository(manifest, first)
build_evaluation_repository(manifest, second)
assert (first / "README.md").read_bytes() == (second / "README.md").read_bytes()
assert (first / ".gitattributes").read_bytes() == (second / ".gitattributes").read_bytes()
metadata = _card_header(first / "README.md")
assert metadata["configs"] == [
{
"config_name": "videomme",
"data_files": [
{
"split": "test",
"path": "annotations/video_qa/videomme/test.jsonl",
}
],
}
]
assert metadata["orarl"]["totals"]["row_count"] == 1
assert metadata["orarl"]["totals"]["asset_count"] == 2
assert metadata["orarl"]["benchmarks"][0]["task"] == "videomme"
assert metadata["orarl"]["benchmarks"][0]["family"] == "video_qa"
assert metadata["size_categories"] == ["n<1K"]
assert "evaluation" in metadata["tags"]
card = (first / "README.md").read_text(encoding="utf-8")
assert "# OraRL-Eval" in card
assert "> **Evaluation only.**" in card
assert "## Benchmark configs and provenance" in card
assert "## Canonical row schema" in card
assert "## Intended use and limitations" in card
assert "arXiv preprint arXiv:2608.20492" in card
assert "load_dataset(" in card
assert "orarl-eval \\" in card
assert "--dataset OraRL/OraRL-Eval" in card
attributes = (first / ".gitattributes").read_text(encoding="utf-8")
assert "annotations/video_qa/videomme/test.jsonl filter=lfs" not in attributes
assert "media/video_qa/videomme/videos/** filter=lfs" in attributes
def test_metadata_cli_refreshes_and_validates_existing_repository(tmp_path: Path) -> None:
root = tmp_path / "repository"
build_evaluation_repository(_source_fixture(tmp_path), root)
expected_readme = (root / "README.md").read_bytes()
(root / "README.md").write_text("stale card\n", encoding="utf-8")
assert eval_data.main(["metadata", "--root", str(root)]) == 0
assert (root / "README.md").read_bytes() == expected_readme
def test_export_index_keeps_only_portable_jsonl_metadata(
tmp_path: Path,
monkeypatch: pytest.MonkeyPatch,
) -> None:
complete = tmp_path / "complete"
index = tmp_path / "index"
fast_index = tmp_path / "fast-index"
build_evaluation_repository(_source_fixture(tmp_path), complete)
summary = export_evaluation_index(complete, index)
def _fail_hash(_path: Path) -> str:
pytest.fail("checksum-free index export must not calculate SHA-256")
monkeypatch.setattr(evaluation_staging, "sha256_file", _fail_hash)
fast_summary = export_evaluation_index(
complete,
fast_index,
checksums=False,
)
assert summary["annotation_assets"] == 1
assert summary["external_assets"] == 1
assert summary["asset_checksums"] is False
assert fast_summary["asset_checksums"] is False
assert (fast_index / "datasets.jsonl").read_bytes() == (
index / "datasets.jsonl"
).read_bytes()
assert (index / "datasets.jsonl").is_file()
assert not (index / "assets.jsonl").exists()
assert (index / "annotations/video_qa/videomme/test.jsonl").is_file()
assert not (index / "media").exists()
assert not (index / "artifacts").exists()
index_card = (index / "README.md").read_text(encoding="utf-8")
assert "pretty_name: OraRL-Data" in index_card
assert "No media is redistributed" in index_card
assert "--asset-root" in index_card
assert index.stat().st_mode & 0o777 == 0o755
assert (index / "datasets.jsonl").stat().st_mode & 0o777 == 0o644
with pytest.raises(FileExistsError, match="refusing to overwrite"):
export_evaluation_index(complete, index)
def test_export_public_keeps_raw_media_and_removes_processed_artifacts(
tmp_path: Path,
) -> None:
complete = tmp_path / "complete"
public = tmp_path / "public"
build_evaluation_repository(
_source_fixture(tmp_path, with_artifact=True),
complete,
)
generated_frame = (
complete
/ "media/video_qa/videomme/videos/test/sam2-cache/000000.jpg"
)
generated_frame.parent.mkdir(parents=True)
generated_frame.write_bytes(b"unmanifested-generated-frame")
summary = export_public_evaluation_repository(complete, public)
assert summary["excluded_artifacts"] == 1
assert (public / "media/video_qa/videomme/videos/test/clip.mp4").is_file()
assert not (public / generated_frame.relative_to(complete)).exists()
assert not (public / "artifacts").exists()
assets = [
json.loads(line)
for line in (public / "assets.jsonl").read_text(encoding="utf-8").splitlines()
]
assert all(record["kind"] != "artifacts" for record in assets)
datasets = [
json.loads(line)
for line in (public / "datasets.jsonl").read_text(encoding="utf-8").splitlines()
]
assert datasets[0]["artifact_paths"] == []
rows = [
json.loads(line)
for line in (
public / "annotations/video_qa/videomme/test.jsonl"
).read_text(encoding="utf-8").splitlines()
]
assert "preprocessed" not in rows[0]
card = (public / "README.md").read_text(encoding="utf-8")
assert "--dataset ./OraRL-Data/OraRL-eval-data" in card
assert "# OraRL-Data" in card
assert "referenced raw media" in card
assert "Preprocessed tensors" in card
assert "No media is redistributed" not in card
assert "path: OraRL-eval-data/annotations/video_qa/videomme/test.jsonl" in card
assert "OraRL-eval-data/media/video_qa/videomme/videos/**" in (
public / ".gitattributes"
).read_text(encoding="utf-8")
(public / "README.md").write_text("stale card\n", encoding="utf-8")
assert (
eval_data.main(
[
"metadata",
"--root",
str(public),
"--repo-id",
"OraRL/OraRL-Data",
]
)
== 0
)
assert "referenced raw media" in (public / "README.md").read_text(
encoding="utf-8"
)
def test_card_subset_and_metadata_apis_group_splits() -> None:
records = [
{
"schema_version": 1,
"benchmark": "spatial_grounding",
"split": split,
"task": "spatial_grounding",
"family": "spatial_grounding",
"annotation_path": f"annotations/spatial_grounding/{split}.jsonl",
"media_paths": ["media/spatial_grounding/images"],
"artifact_paths": [],
"expected_count": count,
"license": "fixture-only",
"source_url": "https://example.org/spatial",
"redistribution_authorized": True,
"evaluation": {"metric_profile": "iou"},
}
for split, count in (("refcoco_val", 2), ("refcocog_test", 3))
]
assets = [
{
"path": f"annotations/spatial_grounding/{split}.jsonl",
"bytes": count * 10,
"kind": "annotations",
"benchmark": "spatial_grounding",
}
for split, count in (("refcoco_val", 2), ("refcocog_test", 3))
]
assert dataset_card_subsets(records) == [
{
"config_name": "spatial_grounding",
"data_files": [
{
"split": "refcoco_val",
"path": "annotations/spatial_grounding/refcoco_val.jsonl",
},
{
"split": "refcocog_test",
"path": "annotations/spatial_grounding/refcocog_test.jsonl",
},
],
}
]
metadata = dataset_card_metadata(records, assets)
assert metadata["orarl"]["totals"] == {
"benchmark_count": 1,
"split_count": 2,
"row_count": 5,
"asset_count": 2,
"byte_count": 50,
}
assert metadata["size_categories"] == ["n<1K"]
assert metadata["tags"][:3] == ["video", "multimodal", "evaluation"]
fields = {item["name"]: item for item in metadata["orarl"]["schema_fields"]}
assert fields["sample_id"]["required"] is True
assert fields["task_payload"]["type"] == "object"
def test_inventory_can_lock_null_expected_counts_for_build(tmp_path: Path) -> None:
manifest = _source_fixture(tmp_path, expected_count=None)
inventory = inventory_evaluation_sources(manifest)
assert inventory["sources"][0]["expected_count"] is None
assert inventory["sources"][0]["count_matches"] is None
assert inventory["totals"]["unlocked_sources"] == 1
with pytest.raises(StagingError, match="expected_count is not locked"):
build_evaluation_repository(manifest, tmp_path / "unlocked-output")
locked = tmp_path / "locked-sources.jsonl"
assert (
eval_data.main(
[
"inventory",
"--manifest",
str(manifest),
"--write-locked-manifest",
str(locked),
"--task",
"videomme",
]
)
== 0
)
locked_record = json.loads(locked.read_text(encoding="utf-8"))
assert locked_record["expected_count"] == 1
assert Path(locked_record["annotation_input"]).is_absolute()
build_evaluation_repository(locked, tmp_path / "locked-output")
with pytest.raises(StagingError, match="no records for task"):
inventory_evaluation_sources(manifest, tasks=["temporal_grounding"])
def test_upload_validates_then_passes_only_noncredential_arguments(
tmp_path: Path,
monkeypatch: pytest.MonkeyPatch,
) -> None:
root = tmp_path / "repository"
build_evaluation_repository(_source_fixture(tmp_path), root)
calls: list[dict[str, object]] = []
constructors: list[tuple[object, ...]] = []
module = types.ModuleType("huggingface_hub")
class FakeApi:
def __init__(self, *args: object, **kwargs: object) -> None:
constructors.append((*args, kwargs))
def upload_large_folder(self, **kwargs: object) -> None:
calls.append(dict(kwargs))
module.HfApi = FakeApi # type: ignore[attr-defined]
monkeypatch.setitem(sys.modules, "huggingface_hub", module)
monkeypatch.setenv("HF_TOKEN", "environment-login-only")
result = upload_evaluation_repository(
root,
"OraRL/OraRL-Eval",
revision="release",
private=True,
num_workers=7,
)
assert constructors == [({},)]
assert calls == [
{
"repo_id": "OraRL/OraRL-Eval",
"folder_path": str(root.resolve()),
"repo_type": "dataset",
"revision": "release",
"private": True,
"num_workers": 7,
}
]
assert "token" not in calls[0]
assert result["rows"] == 1
def test_upload_failure_does_not_echo_authentication_details(
tmp_path: Path,
monkeypatch: pytest.MonkeyPatch,
) -> None:
root = tmp_path / "repository"
build_evaluation_repository(_source_fixture(tmp_path), root)
detail = "sensitive-auth-detail"
module = types.ModuleType("huggingface_hub")
class FakeApi:
def upload_large_folder(self, **_kwargs: object) -> None:
raise RuntimeError(f"authorization failed for {detail}")
module.HfApi = FakeApi # type: ignore[attr-defined]
monkeypatch.setitem(sys.modules, "huggingface_hub", module)
with pytest.raises(UploadError) as captured:
upload_evaluation_repository(root, "OraRL/OraRL-Eval")
assert detail not in str(captured.value)
def test_upload_never_calls_hf_when_full_validation_fails(
tmp_path: Path,
monkeypatch: pytest.MonkeyPatch,
) -> None:
root = tmp_path / "repository"
build_evaluation_repository(_source_fixture(tmp_path), root)
video = next(
path
for path in (root / "media/video_qa/videomme/videos").rglob("*")
if path.is_file()
)
video.write_bytes(b"tampered")
module = types.ModuleType("huggingface_hub")
class ForbiddenApi:
def __init__(self) -> None:
raise AssertionError("network client must not be constructed")
module.HfApi = ForbiddenApi # type: ignore[attr-defined]
monkeypatch.setitem(sys.modules, "huggingface_hub", module)
with pytest.raises(StagingError, match="mismatch"):
upload_evaluation_repository(root, "OraRL/OraRL-Eval")
def test_upload_cli_has_no_token_argument() -> None:
parser = eval_data.create_parser()
with pytest.raises(SystemExit):
parser.parse_args(
[
"upload",
"--root",
"/tmp/repository",
"--repo-id",
"org/repo",
"--token",
"forbidden",
]
)
|