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: 3,143 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 | """Guard the bundled trainer against packaging drift and legacy naming."""
from __future__ import annotations
import ast
from pathlib import Path
from typing import Iterator
import tomllib
import yaml
RELEASE_ROOT = Path(__file__).resolve().parents[1]
RUNTIME_ROOT = RELEASE_ROOT / "verl"
# Internal vocabulary that must not resurface in the public runtime.
FORBIDDEN_RUNTIME_TERMS = (
"".join(("c", "p", "p", "o")),
"".join(("g", "t", "p", "o")),
"".join(("g", "t", "_")),
"".join(("lu", "ff", "y")),
"easy_r1",
)
def _runtime_sources() -> Iterator[Path]:
return (path for path in sorted(RUNTIME_ROOT.rglob("*.py")))
def test_runtime_ships_with_the_release() -> None:
assert (RUNTIME_ROOT / "trainer" / "main.py").is_file()
assert (RUNTIME_ROOT / "trainer" / "ray_trainer.py").is_file()
assert (RUNTIME_ROOT / "workers" / "fsdp_workers.py").is_file()
pyproject = tomllib.loads((RELEASE_ROOT / "pyproject.toml").read_text(encoding="utf-8"))
assert "verl*" in pyproject["tool"]["setuptools"]["packages"]["find"]["include"]
manifest = (RELEASE_ROOT / "MANIFEST.in").read_text(encoding="utf-8")
assert "recursive-include verl *.py" in manifest
def test_orarl_recipes_declare_every_oracle_stage() -> None:
algorithm_config = ast.parse(
(RUNTIME_ROOT / "trainer" / "config.py").read_text(encoding="utf-8")
)
declared = {
statement.target.id
for node in ast.walk(algorithm_config)
if isinstance(node, ast.ClassDef) and node.name == "AlgorithmConfig"
for statement in node.body
if isinstance(statement, ast.AnnAssign) and isinstance(statement.target, ast.Name)
}
for name in ("orarl_4b.yaml", "orarl_9b.yaml"):
payload = yaml.safe_load((RELEASE_ROOT / "configs" / name).read_text(encoding="utf-8"))
algorithm = payload["algorithm"]
assert algorithm["oracle_builder"].startswith("orarl.rewards:")
assert payload["worker"]["reward"]["reward_function"].startswith("orarl.rewards:")
for stage in (
"oracle_injection",
"directional_gain",
"detached_oracle_advantage",
"selection_prune_ratio",
"post_selection_recenter",
):
assert stage in algorithm
assert stage in declared
def test_runtime_uses_the_public_vocabulary() -> None:
offenders: list[str] = []
for path in _runtime_sources():
lowered = path.read_text(encoding="utf-8").casefold()
for term in FORBIDDEN_RUNTIME_TERMS:
if term in lowered:
offenders.append(f"{path.relative_to(RELEASE_ROOT)}: {term}")
assert not offenders, offenders
def test_trainer_entry_point_is_the_bundled_module() -> None:
launcher = (RELEASE_ROOT / "orarl" / "cli" / "train.py").read_text(encoding="utf-8")
assert '"verl.trainer.main"' in launcher
main_source = (RUNTIME_ROOT / "trainer" / "main.py").read_text(encoding="utf-8")
tree = ast.parse(main_source)
functions = {node.name for node in ast.walk(tree) if isinstance(node, ast.FunctionDef)}
assert "main" in functions
|