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: 4,953 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 | # Copyright 2026 The OraRL Authors
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Typed configuration for the dependency-light algorithm helpers."""
from __future__ import annotations
import math
from dataclasses import dataclass, field
def _positive_finite(name: str, value: float) -> None:
if not math.isfinite(value) or value <= 0.0:
raise ValueError(f"{name} must be finite and positive, got {value}.")
def _nonnegative_finite(name: str, value: float) -> None:
if not math.isfinite(value) or value < 0.0:
raise ValueError(f"{name} must be finite and non-negative, got {value}.")
@dataclass(frozen=True, slots=True)
class GRPOConfig:
"""Configuration for the unmodified group-relative baseline."""
normalize: bool = True
eps: float = 1e-6
def __post_init__(self) -> None:
_positive_finite("eps", self.eps)
@dataclass(frozen=True, slots=True)
class DirectionalGainConfig:
"""Configuration for OraRL's positive-direction utility transform."""
gamma: float = 0.25
eps: float = 1e-6
def __post_init__(self) -> None:
_nonnegative_finite("gamma", self.gamma)
_positive_finite("eps", self.eps)
@dataclass(frozen=True, slots=True)
class DetachedOracleConfig:
"""Configuration for the detached annotation advantage."""
scale: float = 2.0
gap_beta: float = 2.0
match_best_ratio: float | None = 1.2
match_best_min: float = 0.05
match_best_max: float = 1.0
eps: float = 1e-6
def __post_init__(self) -> None:
_nonnegative_finite("scale", self.scale)
_positive_finite("gap_beta", self.gap_beta)
_positive_finite("eps", self.eps)
_nonnegative_finite("match_best_min", self.match_best_min)
_nonnegative_finite("match_best_max", self.match_best_max)
if self.match_best_min > self.match_best_max:
raise ValueError("match_best_min must not exceed match_best_max.")
if self.match_best_ratio is not None:
_positive_finite("match_best_ratio", self.match_best_ratio)
@dataclass(frozen=True, slots=True)
class SelectionConfig:
"""Strict sign-balanced selection settings.
The annotation row occupies one slot in ``keep_per_group``. The remaining
slots are assigned to the positive and negative policy quotas.
"""
keep_per_group: int = 4
positive_quota: int = 1
negative_quota: int = 2
world_size: int = 1
def __post_init__(self) -> None:
if self.keep_per_group < 2:
raise ValueError("keep_per_group must leave room for policy and oracle rows.")
if self.positive_quota < 0 or self.negative_quota < 0:
raise ValueError("selection quotas must be non-negative.")
if self.positive_quota + self.negative_quota + 1 != self.keep_per_group:
raise ValueError(
"positive_quota + negative_quota + one oracle slot must equal keep_per_group."
)
if self.world_size <= 0:
raise ValueError("world_size must be positive.")
@dataclass(frozen=True, slots=True)
class CorrectionConfig:
"""Settings for post-selection mean and RMS correction."""
rms_match: bool = True
rms_min_scale: float = 0.25
sigma_policy_floor: float = 1e-3
eps: float = 1e-6
def __post_init__(self) -> None:
if not 0.0 <= self.rms_min_scale <= 1.0:
raise ValueError("rms_min_scale must be in [0, 1].")
_nonnegative_finite("sigma_policy_floor", self.sigma_policy_floor)
_positive_finite("eps", self.eps)
@dataclass(frozen=True, slots=True)
class PostSelectionReference:
"""Pre-selection statistics for one rollout group."""
policy_rms: float
sigma_policy: float
policy_rows: int
def __post_init__(self) -> None:
_nonnegative_finite("policy_rms", self.policy_rms)
_nonnegative_finite("sigma_policy", self.sigma_policy)
if self.policy_rows < 1:
raise ValueError("policy_rows must be positive.")
@dataclass(frozen=True, slots=True)
class OraRLConfig:
"""Paper recipe defaults for the four numerical stages."""
directional_gain: DirectionalGainConfig = field(default_factory=DirectionalGainConfig)
detached_oracle: DetachedOracleConfig = field(default_factory=DetachedOracleConfig)
selection: SelectionConfig = field(default_factory=SelectionConfig)
correction: CorrectionConfig = field(default_factory=CorrectionConfig)
|