For AI agents: a documentation index is available at the root level at /llms.txt and /llms-full.txt. Append /llms.txt to any URL for a page-level index, or .md for the markdown version of any page.
DocumentationAPI Reference
DocumentationAPI Reference
  • API Reference
    • Overview
        • Nemo Curator
          • Backends
          • Config
          • Core
          • Metrics
          • Models
            • Aesthetics
            • Base
            • Client
            • Clip
            • Cosmos Embed1
            • Nemotron 3 Nano Omni
            • Nemotron H Vl
            • Nsfw
            • Prompt Formatter
            • Qwen Lm
            • Qwen Vl
            • Transnetv2
            • Vllm Model
          • Package Info
          • Pipeline
          • Stages
          • Tasks
          • Utils
    • Pipeline
    • ProcessingStage
    • CompositeStage
    • Resources
NVIDIANVIDIA
Developer-friendly docs for your API
Privacy Policy | Your Privacy Choices | Terms of Service | Accessibility | Corporate Policies | Product Security | Contact

Copyright © 2026, NVIDIA Corporation.

LogoLogoNeMo Curator
On this page
  • Module Contents
  • Classes
  • Data
  • API
API ReferenceFull Library ReferenceNemo CuratorNemo CuratorModels

nemo_curator.models.nemotron_3_nano_omni

||View as Markdown|
Previous

nemo_curator.models.cosmos_embed1

Next

nemo_curator.models.nemotron_h_vl

Module Contents

Classes

NameDescription
LLM-
Nemotron3NanoOmniNemotron 3 Nano Omni multimodal VLM for video captioning.
SamplingParams-

Data

VLLM_AVAILABLE

_EXPECTED_VIDEO_TAG_PARTS

_HF_MODEL_ID

_HF_REVISION

_VIDEO_TAG_SPLIT_MAX

API

class nemo_curator.models.nemotron_3_nano_omni.LLM()
class nemo_curator.models.nemotron_3_nano_omni.Nemotron3NanoOmni(
model_dir: str,
caption_batch_size: int = 8,
max_output_tokens: int = 512,
stage2_prompt_text: str | None = None,
verbose: bool = False
)

Bases: ModelInterface

Nemotron 3 Nano Omni multimodal VLM for video captioning.

Weights are downloaded automatically from HuggingFace on first use. model_dir is the base directory; weights are stored at model_dir/<_HF_MODEL_ID>/.

model_id_names
list[str]
stage2_prompt
weight_file
= str(Path(model_dir) / _HF_MODEL_ID)
nemo_curator.models.nemotron_3_nano_omni.Nemotron3NanoOmni._patch_config(
model_dir_path: pathlib.Path
) -> None
staticmethod
nemo_curator.models.nemotron_3_nano_omni.Nemotron3NanoOmni._refine_caption_prompt(
original_prompt: str,
refinement_text: str
) -> str
nemo_curator.models.nemotron_3_nano_omni.Nemotron3NanoOmni.download_weights_on_node(
model_dir: str
) -> None
classmethod

Download Nemotron3NanoOmni weights.

nemo_curator.models.nemotron_3_nano_omni.Nemotron3NanoOmni.generate(
videos: list[dict[str, typing.Any]],
generate_stage2_caption: bool = False,
batch_size: int | None = None
) -> list[str]
nemo_curator.models.nemotron_3_nano_omni.Nemotron3NanoOmni.setup() -> None
class nemo_curator.models.nemotron_3_nano_omni.SamplingParams()
nemo_curator.models.nemotron_3_nano_omni.VLLM_AVAILABLE = True
nemo_curator.models.nemotron_3_nano_omni._EXPECTED_VIDEO_TAG_PARTS = 2
nemo_curator.models.nemotron_3_nano_omni._HF_MODEL_ID: Final[str] = 'nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning'
nemo_curator.models.nemotron_3_nano_omni._HF_REVISION: Final[str] = '23d21acd455d9836d50c48570a329bde77e08ba4'
nemo_curator.models.nemotron_3_nano_omni._VIDEO_TAG_SPLIT_MAX = 1