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: 10,788 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 | #!/usr/bin/env python3
"""Fail closed when the OraRL public tree contains release hygiene hazards."""
from __future__ import annotations
import argparse
import os
import re
from dataclasses import dataclass
from pathlib import Path
from typing import Iterable, Sequence
DEFAULT_MAX_BYTES = 5 * 1024 * 1024
EVALUATION_JSONL_MAX_BYTES = 64 * 1024 * 1024
_METHOD_TERMS = (
"".join(("c", "p", "p", "o")),
"".join(("g", "t", "p", "o")),
"".join(("build_", "g", "t")),
"".join(("lu", "ff", "y")),
)
# Benchmark evaluators name upstream annotation fields after ground truth, so
# this prefix is only a tripwire outside the vendored evaluation runtime.
_LEGACY_ORACLE_TERMS = ("".join(("g", "t", "_")),)
_EXCLUDED_TERMS = _METHOD_TERMS + _LEGACY_ORACLE_TERMS
_IGNORED_LOCAL_DIRECTORIES = {".git", "checkpoint"}
_GENERATED_DIRECTORIES = {
".nox",
".tox",
".venv",
".mypy_cache",
".pytest_cache",
".ruff_cache",
"__pycache__",
"artifacts",
"build",
"cache",
"checkpoints",
"dist",
"htmlcov",
"logs",
"outputs",
"runs",
"venv",
}
_GENERATED_FILES = {".coverage", "orarl-eval-summary.json"}
_ALLOWED_RELEASE_BINARIES = {
Path("assets/orarl-data-scaling.gif"),
Path("assets/orarl-hero.gif"),
Path("assets/orarl-method.gif"),
Path("assets/orarl-model-scaling.gif"),
Path("assets/orarl-teaser.mp4"),
Path("assets/paper-data-scaling.png"),
Path("assets/paper-framework.png"),
Path("assets/paper-model-scaling.png"),
Path("assets/paper-results.png"),
Path("orarl.pdf"),
}
_ARTIFACT_SUFFIXES = {
".arrow",
".avi",
".bin",
".ckpt",
".jsonl",
".mkv",
".mov",
".mp4",
".npy",
".npz",
".parquet",
".pt",
".pth",
".pyc",
".safetensors",
".so",
".webm",
}
_RELEASE_BINARY_FILES = {
Path("assets/orarl-data-scaling.gif"),
Path("assets/orarl-hero.gif"),
Path("assets/orarl-method.gif"),
Path("assets/orarl-model-scaling.gif"),
Path("assets/orarl-teaser.mp4"),
Path("assets/paper-data-scaling.png"),
Path("assets/paper-framework.png"),
Path("assets/paper-model-scaling.png"),
Path("assets/paper-results.png"),
Path("orarl.pdf"),
}
_PRIVATE_PATH_PATTERNS = (
re.compile("/(?:" + "|".join(("mnt", "home", "Users")) + r")/[A-Za-z0-9._-]+(?:/[^\s'\"`]+)?"),
re.compile(
"/" + "data" + "/(?:" + "|".join(("home", "user", "users")) + r")/"
r"[A-Za-z0-9._-]+(?:/[^\s'\"`]+)?"
),
re.compile(
"/(?:" + "|".join(("apd" + "cephfs" + r"[^/\s]*", "jizhi" + "cfs")) + r")(?:/[^\s'\"`]+)?"
),
re.compile("/" + "root" + r"(?:/[^\s'\"`]+)?"),
)
_SECRET_PATTERNS = (
re.compile(
r"(?i)\b(?:api[_-]?key|access[_-]?token|auth[_-]?token|password|secret)"
r"\b\s*[:=]\s*['\"]?[A-Za-z0-9_./+=-]{12,}"
),
re.compile(r"\bAKIA[0-9A-Z]{16}\b"),
re.compile(r"\bgh[pousr]_[A-Za-z0-9]{20,}\b"),
re.compile(r"\bhf_[A-Za-z0-9]{20,}\b"),
re.compile(r"://[^/\s:@]+:[^@\s/]+@"),
)
_PRIVATE_KEY_MARKER = "".join(("-" * 5, "BEGIN ", "PRIVATE", " KEY", "-" * 5))
@dataclass(frozen=True, slots=True)
class Finding:
"""One release gate failure."""
path: Path
reason: str
def _line_number(text: str, index: int) -> int:
return text.count("\n", 0, index) + 1
def _text_findings(
path: Path,
text: str,
*,
check_excluded_terms: bool = True,
allow_ground_truth_names: bool = False,
) -> Iterable[Finding]:
if check_excluded_terms:
lowered = text.casefold()
terms = _METHOD_TERMS if allow_ground_truth_names else _EXCLUDED_TERMS
for term in terms:
index = lowered.find(term)
if index >= 0:
yield Finding(
path,
f"excluded method term at line {_line_number(text, index)}",
)
for pattern in _PRIVATE_PATH_PATTERNS:
match = pattern.search(text)
if match:
yield Finding(
path,
f"private absolute path at line {_line_number(text, match.start())}",
)
if _PRIVATE_KEY_MARKER in text:
yield Finding(path, "private key material")
for pattern in _SECRET_PATTERNS:
match = pattern.search(text)
if match:
yield Finding(
path,
f"credential-like value at line {_line_number(text, match.start())}",
)
def _ignored_generated_directory(root: Path, generated_root: Path) -> bool:
"""Return whether root-level gitignore rules exclude generated state."""
ignore_file = root / ".gitignore"
if not ignore_file.is_file():
return False
try:
patterns = {
line.strip()
for line in ignore_file.read_text(encoding="utf-8").splitlines()
if line.strip() and not line.lstrip().startswith(("#", "!"))
}
except OSError:
return False
name = generated_root.name
if name.endswith(".egg-info") and "*.egg-info/" in patterns:
return True
return f"{name}/" in patterns or f"/{name}/" in patterns
def check_release(root: Path, max_bytes: int = DEFAULT_MAX_BYTES) -> list[Finding]:
"""Return every release hygiene problem under ``root``."""
root = root.expanduser().resolve()
findings: list[Finding] = []
reported_generated: set[Path] = set()
for path in sorted(root.rglob("*")):
relative = path.relative_to(root)
relative_text = relative.as_posix()
lowered_parts = tuple(part.casefold() for part in relative.parts)
release_binary = relative in _RELEASE_BINARY_FILES
release_eval_data = lowered_parts[:2] == ("data", "eval")
release_eval_runtime = lowered_parts[:2] == ("eval", "task")
if any(part in _IGNORED_LOCAL_DIRECTORIES for part in lowered_parts):
continue
generated_index = next(
(
index
for index, part in enumerate(lowered_parts)
if part in _GENERATED_DIRECTORIES or part.endswith(".egg-info")
),
None,
)
if generated_index is not None:
generated_root = Path(*relative.parts[: generated_index + 1])
if _ignored_generated_directory(root, generated_root):
continue
if generated_root not in reported_generated:
findings.append(Finding(generated_root, "generated directory"))
reported_generated.add(generated_root)
continue
path_terms = _METHOD_TERMS if release_eval_runtime else _EXCLUDED_TERMS
if any(term in relative_text.casefold() for term in path_terms):
findings.append(Finding(relative, "excluded method term in path"))
if path.is_symlink():
target = os.readlink(path)
if not path.exists():
findings.append(Finding(relative, "broken symbolic link"))
else:
resolved_target = path.resolve()
try:
resolved_target.relative_to(root)
except ValueError:
findings.append(Finding(relative, "symbolic link escapes release tree"))
for pattern in _PRIVATE_PATH_PATTERNS:
if pattern.search(target):
findings.append(Finding(relative, "symbolic link contains a private path"))
break
continue
if path.is_dir():
if (
lowered_parts
and lowered_parts[0] == "data"
and len(lowered_parts) > 1
and not release_eval_data
):
findings.append(Finding(relative, "generated root data directory"))
continue
if not path.is_file():
findings.append(Finding(relative, "unsupported filesystem entry"))
continue
if relative in _ALLOWED_RELEASE_BINARIES:
continue
if path.name.casefold() in _GENERATED_FILES:
findings.append(Finding(relative, "generated output file"))
if (
path.suffix.casefold() in _ARTIFACT_SUFFIXES
and not release_binary
and not (
release_eval_data
and path.suffix.casefold() == ".jsonl"
)
):
findings.append(Finding(relative, "generated or binary artifact"))
size = path.stat().st_size
effective_max_bytes = (
max(max_bytes, EVALUATION_JSONL_MAX_BYTES)
if release_eval_data and path.suffix.casefold() == ".jsonl"
else max_bytes
)
if size > effective_max_bytes and not release_binary:
findings.append(
Finding(
relative,
f"file is too large ({size} bytes; limit {effective_max_bytes})",
)
)
continue
try:
text = path.read_text(encoding="utf-8")
except UnicodeDecodeError:
if not release_binary:
findings.append(Finding(relative, "non-text file"))
continue
findings.extend(
_text_findings(
relative,
text,
check_excluded_terms=not release_eval_data,
allow_ground_truth_names=release_eval_runtime,
)
)
return findings
def create_parser() -> argparse.ArgumentParser:
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument(
"root",
nargs="?",
default=str(Path(__file__).resolve().parents[1]),
help="Release tree to inspect (default: OraRL source root).",
)
parser.add_argument(
"--max-bytes",
type=int,
default=DEFAULT_MAX_BYTES,
help="Maximum permitted size for one source file.",
)
return parser
def main(argv: Sequence[str] | None = None) -> int:
namespace = create_parser().parse_args(argv)
root = Path(namespace.root).expanduser()
if not root.is_dir():
print(f"ERROR: release root is not a directory: {root}")
return 2
if namespace.max_bytes <= 0:
print("ERROR: --max-bytes must be positive")
return 2
findings = check_release(root, namespace.max_bytes)
if findings:
for finding in findings:
print(f"ERROR: {finding.path}: {finding.reason}")
print(f"release hygiene failed with {len(findings)} finding(s)")
return 1
file_count = sum(1 for path in root.rglob("*") if path.is_file())
print(f"release hygiene passed ({file_count} files)")
return 0
if __name__ == "__main__":
raise SystemExit(main())
|